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// 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
18#include "super.h"
19#include "mds_client.h"
20#include "cache.h"
21#include <linux/ceph/decode.h>
22
23/*
24 * Ceph inode operations
25 *
26 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
27 * setattr, etc.), xattr helpers, and helpers for assimilating
28 * metadata returned by the MDS into our cache.
29 *
30 * Also define helpers for doing asynchronous writeback, invalidation,
31 * and truncation for the benefit of those who can't afford to block
32 * (typically because they are in the message handler path).
33 */
34
35static const struct inode_operations ceph_symlink_iops;
36
37static void ceph_inode_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 struct ceph_inode_info *ci = ceph_inode(inode);
45 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
46
47 ci->i_vino = *(struct ceph_vino *)data;
48 inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
49 inode_set_iversion_raw(inode, 0);
50 percpu_counter_inc(&mdsc->metric.total_inodes);
51
52 return 0;
53}
54
55struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
56{
57 struct inode *inode;
58
59 if (ceph_vino_is_reserved(vino))
60 return ERR_PTR(-EREMOTEIO);
61
62 inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare,
63 ceph_set_ino_cb, &vino);
64 if (!inode)
65 return ERR_PTR(-ENOMEM);
66
67 dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode),
68 ceph_vinop(inode), inode, !!(inode->i_state & I_NEW));
69 return inode;
70}
71
72/*
73 * get/constuct snapdir inode for a given directory
74 */
75struct inode *ceph_get_snapdir(struct inode *parent)
76{
77 struct ceph_vino vino = {
78 .ino = ceph_ino(parent),
79 .snap = CEPH_SNAPDIR,
80 };
81 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
82 struct ceph_inode_info *ci = ceph_inode(inode);
83
84 if (IS_ERR(inode))
85 return inode;
86
87 if (!S_ISDIR(parent->i_mode)) {
88 pr_warn_once("bad snapdir parent type (mode=0%o)\n",
89 parent->i_mode);
90 goto err;
91 }
92
93 if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) {
94 pr_warn_once("bad snapdir inode type (mode=0%o)\n",
95 inode->i_mode);
96 goto err;
97 }
98
99 inode->i_mode = parent->i_mode;
100 inode->i_uid = parent->i_uid;
101 inode->i_gid = parent->i_gid;
102 inode->i_mtime = parent->i_mtime;
103 inode->i_ctime = parent->i_ctime;
104 inode->i_atime = parent->i_atime;
105 ci->i_rbytes = 0;
106 ci->i_btime = ceph_inode(parent)->i_btime;
107
108 if (inode->i_state & I_NEW) {
109 inode->i_op = &ceph_snapdir_iops;
110 inode->i_fop = &ceph_snapdir_fops;
111 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
112 unlock_new_inode(inode);
113 }
114
115 return inode;
116err:
117 if ((inode->i_state & I_NEW))
118 discard_new_inode(inode);
119 else
120 iput(inode);
121 return ERR_PTR(-ENOTDIR);
122}
123
124const struct inode_operations ceph_file_iops = {
125 .permission = ceph_permission,
126 .setattr = ceph_setattr,
127 .getattr = ceph_getattr,
128 .listxattr = ceph_listxattr,
129 .get_inode_acl = ceph_get_acl,
130 .set_acl = ceph_set_acl,
131};
132
133
134/*
135 * We use a 'frag tree' to keep track of the MDS's directory fragments
136 * for a given inode (usually there is just a single fragment). We
137 * need to know when a child frag is delegated to a new MDS, or when
138 * it is flagged as replicated, so we can direct our requests
139 * accordingly.
140 */
141
142/*
143 * find/create a frag in the tree
144 */
145static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
146 u32 f)
147{
148 struct rb_node **p;
149 struct rb_node *parent = NULL;
150 struct ceph_inode_frag *frag;
151 int c;
152
153 p = &ci->i_fragtree.rb_node;
154 while (*p) {
155 parent = *p;
156 frag = rb_entry(parent, struct ceph_inode_frag, node);
157 c = ceph_frag_compare(f, frag->frag);
158 if (c < 0)
159 p = &(*p)->rb_left;
160 else if (c > 0)
161 p = &(*p)->rb_right;
162 else
163 return frag;
164 }
165
166 frag = kmalloc(sizeof(*frag), GFP_NOFS);
167 if (!frag)
168 return ERR_PTR(-ENOMEM);
169
170 frag->frag = f;
171 frag->split_by = 0;
172 frag->mds = -1;
173 frag->ndist = 0;
174
175 rb_link_node(&frag->node, parent, p);
176 rb_insert_color(&frag->node, &ci->i_fragtree);
177
178 dout("get_or_create_frag added %llx.%llx frag %x\n",
179 ceph_vinop(&ci->netfs.inode), f);
180 return frag;
181}
182
183/*
184 * find a specific frag @f
185 */
186struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
187{
188 struct rb_node *n = ci->i_fragtree.rb_node;
189
190 while (n) {
191 struct ceph_inode_frag *frag =
192 rb_entry(n, struct ceph_inode_frag, node);
193 int c = ceph_frag_compare(f, frag->frag);
194 if (c < 0)
195 n = n->rb_left;
196 else if (c > 0)
197 n = n->rb_right;
198 else
199 return frag;
200 }
201 return NULL;
202}
203
204/*
205 * Choose frag containing the given value @v. If @pfrag is
206 * specified, copy the frag delegation info to the caller if
207 * it is present.
208 */
209static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
210 struct ceph_inode_frag *pfrag, int *found)
211{
212 u32 t = ceph_frag_make(0, 0);
213 struct ceph_inode_frag *frag;
214 unsigned nway, i;
215 u32 n;
216
217 if (found)
218 *found = 0;
219
220 while (1) {
221 WARN_ON(!ceph_frag_contains_value(t, v));
222 frag = __ceph_find_frag(ci, t);
223 if (!frag)
224 break; /* t is a leaf */
225 if (frag->split_by == 0) {
226 if (pfrag)
227 memcpy(pfrag, frag, sizeof(*pfrag));
228 if (found)
229 *found = 1;
230 break;
231 }
232
233 /* choose child */
234 nway = 1 << frag->split_by;
235 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
236 frag->split_by, nway);
237 for (i = 0; i < nway; i++) {
238 n = ceph_frag_make_child(t, frag->split_by, i);
239 if (ceph_frag_contains_value(n, v)) {
240 t = n;
241 break;
242 }
243 }
244 BUG_ON(i == nway);
245 }
246 dout("choose_frag(%x) = %x\n", v, t);
247
248 return t;
249}
250
251u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
252 struct ceph_inode_frag *pfrag, int *found)
253{
254 u32 ret;
255 mutex_lock(&ci->i_fragtree_mutex);
256 ret = __ceph_choose_frag(ci, v, pfrag, found);
257 mutex_unlock(&ci->i_fragtree_mutex);
258 return ret;
259}
260
261/*
262 * Process dirfrag (delegation) info from the mds. Include leaf
263 * fragment in tree ONLY if ndist > 0. Otherwise, only
264 * branches/splits are included in i_fragtree)
265 */
266static int ceph_fill_dirfrag(struct inode *inode,
267 struct ceph_mds_reply_dirfrag *dirinfo)
268{
269 struct ceph_inode_info *ci = ceph_inode(inode);
270 struct ceph_inode_frag *frag;
271 u32 id = le32_to_cpu(dirinfo->frag);
272 int mds = le32_to_cpu(dirinfo->auth);
273 int ndist = le32_to_cpu(dirinfo->ndist);
274 int diri_auth = -1;
275 int i;
276 int err = 0;
277
278 spin_lock(&ci->i_ceph_lock);
279 if (ci->i_auth_cap)
280 diri_auth = ci->i_auth_cap->mds;
281 spin_unlock(&ci->i_ceph_lock);
282
283 if (mds == -1) /* CDIR_AUTH_PARENT */
284 mds = diri_auth;
285
286 mutex_lock(&ci->i_fragtree_mutex);
287 if (ndist == 0 && mds == diri_auth) {
288 /* no delegation info needed. */
289 frag = __ceph_find_frag(ci, id);
290 if (!frag)
291 goto out;
292 if (frag->split_by == 0) {
293 /* tree leaf, remove */
294 dout("fill_dirfrag removed %llx.%llx frag %x"
295 " (no ref)\n", ceph_vinop(inode), id);
296 rb_erase(&frag->node, &ci->i_fragtree);
297 kfree(frag);
298 } else {
299 /* tree branch, keep and clear */
300 dout("fill_dirfrag cleared %llx.%llx frag %x"
301 " referral\n", ceph_vinop(inode), id);
302 frag->mds = -1;
303 frag->ndist = 0;
304 }
305 goto out;
306 }
307
308
309 /* find/add this frag to store mds delegation info */
310 frag = __get_or_create_frag(ci, id);
311 if (IS_ERR(frag)) {
312 /* this is not the end of the world; we can continue
313 with bad/inaccurate delegation info */
314 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
315 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
316 err = -ENOMEM;
317 goto out;
318 }
319
320 frag->mds = mds;
321 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
322 for (i = 0; i < frag->ndist; i++)
323 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
324 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
325 ceph_vinop(inode), frag->frag, frag->ndist);
326
327out:
328 mutex_unlock(&ci->i_fragtree_mutex);
329 return err;
330}
331
332static int frag_tree_split_cmp(const void *l, const void *r)
333{
334 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
335 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
336 return ceph_frag_compare(le32_to_cpu(ls->frag),
337 le32_to_cpu(rs->frag));
338}
339
340static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
341{
342 if (!frag)
343 return f == ceph_frag_make(0, 0);
344 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
345 return false;
346 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
347}
348
349static int ceph_fill_fragtree(struct inode *inode,
350 struct ceph_frag_tree_head *fragtree,
351 struct ceph_mds_reply_dirfrag *dirinfo)
352{
353 struct ceph_inode_info *ci = ceph_inode(inode);
354 struct ceph_inode_frag *frag, *prev_frag = NULL;
355 struct rb_node *rb_node;
356 unsigned i, split_by, nsplits;
357 u32 id;
358 bool update = false;
359
360 mutex_lock(&ci->i_fragtree_mutex);
361 nsplits = le32_to_cpu(fragtree->nsplits);
362 if (nsplits != ci->i_fragtree_nsplits) {
363 update = true;
364 } else if (nsplits) {
365 i = get_random_u32_below(nsplits);
366 id = le32_to_cpu(fragtree->splits[i].frag);
367 if (!__ceph_find_frag(ci, id))
368 update = true;
369 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
370 rb_node = rb_first(&ci->i_fragtree);
371 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
372 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
373 update = true;
374 }
375 if (!update && dirinfo) {
376 id = le32_to_cpu(dirinfo->frag);
377 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
378 update = true;
379 }
380 if (!update)
381 goto out_unlock;
382
383 if (nsplits > 1) {
384 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
385 frag_tree_split_cmp, NULL);
386 }
387
388 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
389 rb_node = rb_first(&ci->i_fragtree);
390 for (i = 0; i < nsplits; i++) {
391 id = le32_to_cpu(fragtree->splits[i].frag);
392 split_by = le32_to_cpu(fragtree->splits[i].by);
393 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
394 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
395 "frag %x split by %d\n", ceph_vinop(inode),
396 i, nsplits, id, split_by);
397 continue;
398 }
399 frag = NULL;
400 while (rb_node) {
401 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
402 if (ceph_frag_compare(frag->frag, id) >= 0) {
403 if (frag->frag != id)
404 frag = NULL;
405 else
406 rb_node = rb_next(rb_node);
407 break;
408 }
409 rb_node = rb_next(rb_node);
410 /* delete stale split/leaf node */
411 if (frag->split_by > 0 ||
412 !is_frag_child(frag->frag, prev_frag)) {
413 rb_erase(&frag->node, &ci->i_fragtree);
414 if (frag->split_by > 0)
415 ci->i_fragtree_nsplits--;
416 kfree(frag);
417 }
418 frag = NULL;
419 }
420 if (!frag) {
421 frag = __get_or_create_frag(ci, id);
422 if (IS_ERR(frag))
423 continue;
424 }
425 if (frag->split_by == 0)
426 ci->i_fragtree_nsplits++;
427 frag->split_by = split_by;
428 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
429 prev_frag = frag;
430 }
431 while (rb_node) {
432 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
433 rb_node = rb_next(rb_node);
434 /* delete stale split/leaf node */
435 if (frag->split_by > 0 ||
436 !is_frag_child(frag->frag, prev_frag)) {
437 rb_erase(&frag->node, &ci->i_fragtree);
438 if (frag->split_by > 0)
439 ci->i_fragtree_nsplits--;
440 kfree(frag);
441 }
442 }
443out_unlock:
444 mutex_unlock(&ci->i_fragtree_mutex);
445 return 0;
446}
447
448/*
449 * initialize a newly allocated inode.
450 */
451struct inode *ceph_alloc_inode(struct super_block *sb)
452{
453 struct ceph_inode_info *ci;
454 int i;
455
456 ci = alloc_inode_sb(sb, ceph_inode_cachep, GFP_NOFS);
457 if (!ci)
458 return NULL;
459
460 dout("alloc_inode %p\n", &ci->netfs.inode);
461
462 /* Set parameters for the netfs library */
463 netfs_inode_init(&ci->netfs, &ceph_netfs_ops);
464
465 spin_lock_init(&ci->i_ceph_lock);
466
467 ci->i_version = 0;
468 ci->i_inline_version = 0;
469 ci->i_time_warp_seq = 0;
470 ci->i_ceph_flags = 0;
471 atomic64_set(&ci->i_ordered_count, 1);
472 atomic64_set(&ci->i_release_count, 1);
473 atomic64_set(&ci->i_complete_seq[0], 0);
474 atomic64_set(&ci->i_complete_seq[1], 0);
475 ci->i_symlink = NULL;
476
477 ci->i_max_bytes = 0;
478 ci->i_max_files = 0;
479
480 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
481 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
482 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
483
484 ci->i_fragtree = RB_ROOT;
485 mutex_init(&ci->i_fragtree_mutex);
486
487 ci->i_xattrs.blob = NULL;
488 ci->i_xattrs.prealloc_blob = NULL;
489 ci->i_xattrs.dirty = false;
490 ci->i_xattrs.index = RB_ROOT;
491 ci->i_xattrs.count = 0;
492 ci->i_xattrs.names_size = 0;
493 ci->i_xattrs.vals_size = 0;
494 ci->i_xattrs.version = 0;
495 ci->i_xattrs.index_version = 0;
496
497 ci->i_caps = RB_ROOT;
498 ci->i_auth_cap = NULL;
499 ci->i_dirty_caps = 0;
500 ci->i_flushing_caps = 0;
501 INIT_LIST_HEAD(&ci->i_dirty_item);
502 INIT_LIST_HEAD(&ci->i_flushing_item);
503 ci->i_prealloc_cap_flush = NULL;
504 INIT_LIST_HEAD(&ci->i_cap_flush_list);
505 init_waitqueue_head(&ci->i_cap_wq);
506 ci->i_hold_caps_max = 0;
507 INIT_LIST_HEAD(&ci->i_cap_delay_list);
508 INIT_LIST_HEAD(&ci->i_cap_snaps);
509 ci->i_head_snapc = NULL;
510 ci->i_snap_caps = 0;
511
512 ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
513 for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
514 ci->i_nr_by_mode[i] = 0;
515
516 mutex_init(&ci->i_truncate_mutex);
517 ci->i_truncate_seq = 0;
518 ci->i_truncate_size = 0;
519 ci->i_truncate_pending = 0;
520
521 ci->i_max_size = 0;
522 ci->i_reported_size = 0;
523 ci->i_wanted_max_size = 0;
524 ci->i_requested_max_size = 0;
525
526 ci->i_pin_ref = 0;
527 ci->i_rd_ref = 0;
528 ci->i_rdcache_ref = 0;
529 ci->i_wr_ref = 0;
530 ci->i_wb_ref = 0;
531 ci->i_fx_ref = 0;
532 ci->i_wrbuffer_ref = 0;
533 ci->i_wrbuffer_ref_head = 0;
534 atomic_set(&ci->i_filelock_ref, 0);
535 atomic_set(&ci->i_shared_gen, 1);
536 ci->i_rdcache_gen = 0;
537 ci->i_rdcache_revoking = 0;
538
539 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
540 INIT_LIST_HEAD(&ci->i_unsafe_iops);
541 spin_lock_init(&ci->i_unsafe_lock);
542
543 ci->i_snap_realm = NULL;
544 INIT_LIST_HEAD(&ci->i_snap_realm_item);
545 INIT_LIST_HEAD(&ci->i_snap_flush_item);
546
547 INIT_WORK(&ci->i_work, ceph_inode_work);
548 ci->i_work_mask = 0;
549 memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
550 return &ci->netfs.inode;
551}
552
553void ceph_free_inode(struct inode *inode)
554{
555 struct ceph_inode_info *ci = ceph_inode(inode);
556
557 kfree(ci->i_symlink);
558 kmem_cache_free(ceph_inode_cachep, ci);
559}
560
561void ceph_evict_inode(struct inode *inode)
562{
563 struct ceph_inode_info *ci = ceph_inode(inode);
564 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
565 struct ceph_inode_frag *frag;
566 struct rb_node *n;
567
568 dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
569
570 percpu_counter_dec(&mdsc->metric.total_inodes);
571
572 truncate_inode_pages_final(&inode->i_data);
573 if (inode->i_state & I_PINNING_FSCACHE_WB)
574 ceph_fscache_unuse_cookie(inode, true);
575 clear_inode(inode);
576
577 ceph_fscache_unregister_inode_cookie(ci);
578
579 __ceph_remove_caps(ci);
580
581 if (__ceph_has_quota(ci, QUOTA_GET_ANY))
582 ceph_adjust_quota_realms_count(inode, false);
583
584 /*
585 * we may still have a snap_realm reference if there are stray
586 * caps in i_snap_caps.
587 */
588 if (ci->i_snap_realm) {
589 if (ceph_snap(inode) == CEPH_NOSNAP) {
590 dout(" dropping residual ref to snap realm %p\n",
591 ci->i_snap_realm);
592 ceph_change_snap_realm(inode, NULL);
593 } else {
594 ceph_put_snapid_map(mdsc, ci->i_snapid_map);
595 ci->i_snap_realm = NULL;
596 }
597 }
598
599 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
600 frag = rb_entry(n, struct ceph_inode_frag, node);
601 rb_erase(n, &ci->i_fragtree);
602 kfree(frag);
603 }
604 ci->i_fragtree_nsplits = 0;
605
606 __ceph_destroy_xattrs(ci);
607 if (ci->i_xattrs.blob)
608 ceph_buffer_put(ci->i_xattrs.blob);
609 if (ci->i_xattrs.prealloc_blob)
610 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
611
612 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
613 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
614}
615
616static inline blkcnt_t calc_inode_blocks(u64 size)
617{
618 return (size + (1<<9) - 1) >> 9;
619}
620
621/*
622 * Helpers to fill in size, ctime, mtime, and atime. We have to be
623 * careful because either the client or MDS may have more up to date
624 * info, depending on which capabilities are held, and whether
625 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
626 * and size are monotonically increasing, except when utimes() or
627 * truncate() increments the corresponding _seq values.)
628 */
629int ceph_fill_file_size(struct inode *inode, int issued,
630 u32 truncate_seq, u64 truncate_size, u64 size)
631{
632 struct ceph_inode_info *ci = ceph_inode(inode);
633 int queue_trunc = 0;
634 loff_t isize = i_size_read(inode);
635
636 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
637 (truncate_seq == ci->i_truncate_seq && size > isize)) {
638 dout("size %lld -> %llu\n", isize, size);
639 if (size > 0 && S_ISDIR(inode->i_mode)) {
640 pr_err("fill_file_size non-zero size for directory\n");
641 size = 0;
642 }
643 i_size_write(inode, size);
644 inode->i_blocks = calc_inode_blocks(size);
645 /*
646 * If we're expanding, then we should be able to just update
647 * the existing cookie.
648 */
649 if (size > isize)
650 ceph_fscache_update(inode);
651 ci->i_reported_size = size;
652 if (truncate_seq != ci->i_truncate_seq) {
653 dout("truncate_seq %u -> %u\n",
654 ci->i_truncate_seq, truncate_seq);
655 ci->i_truncate_seq = truncate_seq;
656
657 /* the MDS should have revoked these caps */
658 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
659 CEPH_CAP_FILE_RD |
660 CEPH_CAP_FILE_WR |
661 CEPH_CAP_FILE_LAZYIO));
662 /*
663 * If we hold relevant caps, or in the case where we're
664 * not the only client referencing this file and we
665 * don't hold those caps, then we need to check whether
666 * the file is either opened or mmaped
667 */
668 if ((issued & (CEPH_CAP_FILE_CACHE|
669 CEPH_CAP_FILE_BUFFER)) ||
670 mapping_mapped(inode->i_mapping) ||
671 __ceph_is_file_opened(ci)) {
672 ci->i_truncate_pending++;
673 queue_trunc = 1;
674 }
675 }
676 }
677 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
678 ci->i_truncate_size != truncate_size) {
679 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
680 truncate_size);
681 ci->i_truncate_size = truncate_size;
682 }
683 return queue_trunc;
684}
685
686void ceph_fill_file_time(struct inode *inode, int issued,
687 u64 time_warp_seq, struct timespec64 *ctime,
688 struct timespec64 *mtime, struct timespec64 *atime)
689{
690 struct ceph_inode_info *ci = ceph_inode(inode);
691 int warn = 0;
692
693 if (issued & (CEPH_CAP_FILE_EXCL|
694 CEPH_CAP_FILE_WR|
695 CEPH_CAP_FILE_BUFFER|
696 CEPH_CAP_AUTH_EXCL|
697 CEPH_CAP_XATTR_EXCL)) {
698 if (ci->i_version == 0 ||
699 timespec64_compare(ctime, &inode->i_ctime) > 0) {
700 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
701 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
702 ctime->tv_sec, ctime->tv_nsec);
703 inode->i_ctime = *ctime;
704 }
705 if (ci->i_version == 0 ||
706 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
707 /* the MDS did a utimes() */
708 dout("mtime %lld.%09ld -> %lld.%09ld "
709 "tw %d -> %d\n",
710 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
711 mtime->tv_sec, mtime->tv_nsec,
712 ci->i_time_warp_seq, (int)time_warp_seq);
713
714 inode->i_mtime = *mtime;
715 inode->i_atime = *atime;
716 ci->i_time_warp_seq = time_warp_seq;
717 } else if (time_warp_seq == ci->i_time_warp_seq) {
718 /* nobody did utimes(); take the max */
719 if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
720 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
721 inode->i_mtime.tv_sec,
722 inode->i_mtime.tv_nsec,
723 mtime->tv_sec, mtime->tv_nsec);
724 inode->i_mtime = *mtime;
725 }
726 if (timespec64_compare(atime, &inode->i_atime) > 0) {
727 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
728 inode->i_atime.tv_sec,
729 inode->i_atime.tv_nsec,
730 atime->tv_sec, atime->tv_nsec);
731 inode->i_atime = *atime;
732 }
733 } else if (issued & CEPH_CAP_FILE_EXCL) {
734 /* we did a utimes(); ignore mds values */
735 } else {
736 warn = 1;
737 }
738 } else {
739 /* we have no write|excl caps; whatever the MDS says is true */
740 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
741 inode->i_ctime = *ctime;
742 inode->i_mtime = *mtime;
743 inode->i_atime = *atime;
744 ci->i_time_warp_seq = time_warp_seq;
745 } else {
746 warn = 1;
747 }
748 }
749 if (warn) /* time_warp_seq shouldn't go backwards */
750 dout("%p mds time_warp_seq %llu < %u\n",
751 inode, time_warp_seq, ci->i_time_warp_seq);
752}
753
754/*
755 * Populate an inode based on info from mds. May be called on new or
756 * existing inodes.
757 */
758int ceph_fill_inode(struct inode *inode, struct page *locked_page,
759 struct ceph_mds_reply_info_in *iinfo,
760 struct ceph_mds_reply_dirfrag *dirinfo,
761 struct ceph_mds_session *session, int cap_fmode,
762 struct ceph_cap_reservation *caps_reservation)
763{
764 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
765 struct ceph_mds_reply_inode *info = iinfo->in;
766 struct ceph_inode_info *ci = ceph_inode(inode);
767 int issued, new_issued, info_caps;
768 struct timespec64 mtime, atime, ctime;
769 struct ceph_buffer *xattr_blob = NULL;
770 struct ceph_buffer *old_blob = NULL;
771 struct ceph_string *pool_ns = NULL;
772 struct ceph_cap *new_cap = NULL;
773 int err = 0;
774 bool wake = false;
775 bool queue_trunc = false;
776 bool new_version = false;
777 bool fill_inline = false;
778 umode_t mode = le32_to_cpu(info->mode);
779 dev_t rdev = le32_to_cpu(info->rdev);
780
781 lockdep_assert_held(&mdsc->snap_rwsem);
782
783 dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
784 inode, ceph_vinop(inode), le64_to_cpu(info->version),
785 ci->i_version);
786
787 /* Once I_NEW is cleared, we can't change type or dev numbers */
788 if (inode->i_state & I_NEW) {
789 inode->i_mode = mode;
790 } else {
791 if (inode_wrong_type(inode, mode)) {
792 pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
793 ceph_vinop(inode), inode->i_mode, mode);
794 return -ESTALE;
795 }
796
797 if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
798 pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
799 ceph_vinop(inode), MAJOR(inode->i_rdev),
800 MINOR(inode->i_rdev), MAJOR(rdev),
801 MINOR(rdev));
802 return -ESTALE;
803 }
804 }
805
806 info_caps = le32_to_cpu(info->cap.caps);
807
808 /* prealloc new cap struct */
809 if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
810 new_cap = ceph_get_cap(mdsc, caps_reservation);
811 if (!new_cap)
812 return -ENOMEM;
813 }
814
815 /*
816 * prealloc xattr data, if it looks like we'll need it. only
817 * if len > 4 (meaning there are actually xattrs; the first 4
818 * bytes are the xattr count).
819 */
820 if (iinfo->xattr_len > 4) {
821 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
822 if (!xattr_blob)
823 pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
824 iinfo->xattr_len);
825 }
826
827 if (iinfo->pool_ns_len > 0)
828 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
829 iinfo->pool_ns_len);
830
831 if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
832 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
833
834 spin_lock(&ci->i_ceph_lock);
835
836 /*
837 * provided version will be odd if inode value is projected,
838 * even if stable. skip the update if we have newer stable
839 * info (ours>=theirs, e.g. due to racing mds replies), unless
840 * we are getting projected (unstable) info (in which case the
841 * version is odd, and we want ours>theirs).
842 * us them
843 * 2 2 skip
844 * 3 2 skip
845 * 3 3 update
846 */
847 if (ci->i_version == 0 ||
848 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
849 le64_to_cpu(info->version) > (ci->i_version & ~1)))
850 new_version = true;
851
852 /* Update change_attribute */
853 inode_set_max_iversion_raw(inode, iinfo->change_attr);
854
855 __ceph_caps_issued(ci, &issued);
856 issued |= __ceph_caps_dirty(ci);
857 new_issued = ~issued & info_caps;
858
859 /* directories have fl_stripe_unit set to zero */
860 if (le32_to_cpu(info->layout.fl_stripe_unit))
861 inode->i_blkbits =
862 fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
863 else
864 inode->i_blkbits = CEPH_BLOCK_SHIFT;
865
866 __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
867
868 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
869 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
870 inode->i_mode = mode;
871 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
872 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
873 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
874 from_kuid(&init_user_ns, inode->i_uid),
875 from_kgid(&init_user_ns, inode->i_gid));
876 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
877 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
878 }
879
880 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
881 (issued & CEPH_CAP_LINK_EXCL) == 0)
882 set_nlink(inode, le32_to_cpu(info->nlink));
883
884 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
885 /* be careful with mtime, atime, size */
886 ceph_decode_timespec64(&atime, &info->atime);
887 ceph_decode_timespec64(&mtime, &info->mtime);
888 ceph_decode_timespec64(&ctime, &info->ctime);
889 ceph_fill_file_time(inode, issued,
890 le32_to_cpu(info->time_warp_seq),
891 &ctime, &mtime, &atime);
892 }
893
894 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
895 ci->i_files = le64_to_cpu(info->files);
896 ci->i_subdirs = le64_to_cpu(info->subdirs);
897 }
898
899 if (new_version ||
900 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
901 s64 old_pool = ci->i_layout.pool_id;
902 struct ceph_string *old_ns;
903
904 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
905 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
906 lockdep_is_held(&ci->i_ceph_lock));
907 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
908
909 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
910 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
911
912 pool_ns = old_ns;
913
914 queue_trunc = ceph_fill_file_size(inode, issued,
915 le32_to_cpu(info->truncate_seq),
916 le64_to_cpu(info->truncate_size),
917 le64_to_cpu(info->size));
918 /* only update max_size on auth cap */
919 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
920 ci->i_max_size != le64_to_cpu(info->max_size)) {
921 dout("max_size %lld -> %llu\n", ci->i_max_size,
922 le64_to_cpu(info->max_size));
923 ci->i_max_size = le64_to_cpu(info->max_size);
924 }
925 }
926
927 /* layout and rstat are not tracked by capability, update them if
928 * the inode info is from auth mds */
929 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
930 if (S_ISDIR(inode->i_mode)) {
931 ci->i_dir_layout = iinfo->dir_layout;
932 ci->i_rbytes = le64_to_cpu(info->rbytes);
933 ci->i_rfiles = le64_to_cpu(info->rfiles);
934 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
935 ci->i_dir_pin = iinfo->dir_pin;
936 ci->i_rsnaps = iinfo->rsnaps;
937 ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
938 }
939 }
940
941 /* xattrs */
942 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
943 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
944 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
945 if (ci->i_xattrs.blob)
946 old_blob = ci->i_xattrs.blob;
947 ci->i_xattrs.blob = xattr_blob;
948 if (xattr_blob)
949 memcpy(ci->i_xattrs.blob->vec.iov_base,
950 iinfo->xattr_data, iinfo->xattr_len);
951 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
952 ceph_forget_all_cached_acls(inode);
953 ceph_security_invalidate_secctx(inode);
954 xattr_blob = NULL;
955 }
956
957 /* finally update i_version */
958 if (le64_to_cpu(info->version) > ci->i_version)
959 ci->i_version = le64_to_cpu(info->version);
960
961 inode->i_mapping->a_ops = &ceph_aops;
962
963 switch (inode->i_mode & S_IFMT) {
964 case S_IFIFO:
965 case S_IFBLK:
966 case S_IFCHR:
967 case S_IFSOCK:
968 inode->i_blkbits = PAGE_SHIFT;
969 init_special_inode(inode, inode->i_mode, rdev);
970 inode->i_op = &ceph_file_iops;
971 break;
972 case S_IFREG:
973 inode->i_op = &ceph_file_iops;
974 inode->i_fop = &ceph_file_fops;
975 break;
976 case S_IFLNK:
977 inode->i_op = &ceph_symlink_iops;
978 if (!ci->i_symlink) {
979 u32 symlen = iinfo->symlink_len;
980 char *sym;
981
982 spin_unlock(&ci->i_ceph_lock);
983
984 if (symlen != i_size_read(inode)) {
985 pr_err("%s %llx.%llx BAD symlink "
986 "size %lld\n", __func__,
987 ceph_vinop(inode),
988 i_size_read(inode));
989 i_size_write(inode, symlen);
990 inode->i_blocks = calc_inode_blocks(symlen);
991 }
992
993 err = -ENOMEM;
994 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
995 if (!sym)
996 goto out;
997
998 spin_lock(&ci->i_ceph_lock);
999 if (!ci->i_symlink)
1000 ci->i_symlink = sym;
1001 else
1002 kfree(sym); /* lost a race */
1003 }
1004 inode->i_link = ci->i_symlink;
1005 break;
1006 case S_IFDIR:
1007 inode->i_op = &ceph_dir_iops;
1008 inode->i_fop = &ceph_dir_fops;
1009 break;
1010 default:
1011 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
1012 ceph_vinop(inode), inode->i_mode);
1013 }
1014
1015 /* were we issued a capability? */
1016 if (info_caps) {
1017 if (ceph_snap(inode) == CEPH_NOSNAP) {
1018 ceph_add_cap(inode, session,
1019 le64_to_cpu(info->cap.cap_id),
1020 info_caps,
1021 le32_to_cpu(info->cap.wanted),
1022 le32_to_cpu(info->cap.seq),
1023 le32_to_cpu(info->cap.mseq),
1024 le64_to_cpu(info->cap.realm),
1025 info->cap.flags, &new_cap);
1026
1027 /* set dir completion flag? */
1028 if (S_ISDIR(inode->i_mode) &&
1029 ci->i_files == 0 && ci->i_subdirs == 0 &&
1030 (info_caps & CEPH_CAP_FILE_SHARED) &&
1031 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1032 !__ceph_dir_is_complete(ci)) {
1033 dout(" marking %p complete (empty)\n", inode);
1034 i_size_write(inode, 0);
1035 __ceph_dir_set_complete(ci,
1036 atomic64_read(&ci->i_release_count),
1037 atomic64_read(&ci->i_ordered_count));
1038 }
1039
1040 wake = true;
1041 } else {
1042 dout(" %p got snap_caps %s\n", inode,
1043 ceph_cap_string(info_caps));
1044 ci->i_snap_caps |= info_caps;
1045 }
1046 }
1047
1048 if (iinfo->inline_version > 0 &&
1049 iinfo->inline_version >= ci->i_inline_version) {
1050 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1051 ci->i_inline_version = iinfo->inline_version;
1052 if (ceph_has_inline_data(ci) &&
1053 (locked_page || (info_caps & cache_caps)))
1054 fill_inline = true;
1055 }
1056
1057 if (cap_fmode >= 0) {
1058 if (!info_caps)
1059 pr_warn("mds issued no caps on %llx.%llx\n",
1060 ceph_vinop(inode));
1061 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1062 }
1063
1064 spin_unlock(&ci->i_ceph_lock);
1065
1066 ceph_fscache_register_inode_cookie(inode);
1067
1068 if (fill_inline)
1069 ceph_fill_inline_data(inode, locked_page,
1070 iinfo->inline_data, iinfo->inline_len);
1071
1072 if (wake)
1073 wake_up_all(&ci->i_cap_wq);
1074
1075 /* queue truncate if we saw i_size decrease */
1076 if (queue_trunc)
1077 ceph_queue_vmtruncate(inode);
1078
1079 /* populate frag tree */
1080 if (S_ISDIR(inode->i_mode))
1081 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1082
1083 /* update delegation info? */
1084 if (dirinfo)
1085 ceph_fill_dirfrag(inode, dirinfo);
1086
1087 err = 0;
1088out:
1089 if (new_cap)
1090 ceph_put_cap(mdsc, new_cap);
1091 ceph_buffer_put(old_blob);
1092 ceph_buffer_put(xattr_blob);
1093 ceph_put_string(pool_ns);
1094 return err;
1095}
1096
1097/*
1098 * caller should hold session s_mutex and dentry->d_lock.
1099 */
1100static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1101 struct ceph_mds_reply_lease *lease,
1102 struct ceph_mds_session *session,
1103 unsigned long from_time,
1104 struct ceph_mds_session **old_lease_session)
1105{
1106 struct ceph_dentry_info *di = ceph_dentry(dentry);
1107 unsigned mask = le16_to_cpu(lease->mask);
1108 long unsigned duration = le32_to_cpu(lease->duration_ms);
1109 long unsigned ttl = from_time + (duration * HZ) / 1000;
1110 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1111
1112 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1113 dentry, duration, ttl);
1114
1115 /* only track leases on regular dentries */
1116 if (ceph_snap(dir) != CEPH_NOSNAP)
1117 return;
1118
1119 if (mask & CEPH_LEASE_PRIMARY_LINK)
1120 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1121 else
1122 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1123
1124 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1125 if (!(mask & CEPH_LEASE_VALID)) {
1126 __ceph_dentry_dir_lease_touch(di);
1127 return;
1128 }
1129
1130 if (di->lease_gen == atomic_read(&session->s_cap_gen) &&
1131 time_before(ttl, di->time))
1132 return; /* we already have a newer lease. */
1133
1134 if (di->lease_session && di->lease_session != session) {
1135 *old_lease_session = di->lease_session;
1136 di->lease_session = NULL;
1137 }
1138
1139 if (!di->lease_session)
1140 di->lease_session = ceph_get_mds_session(session);
1141 di->lease_gen = atomic_read(&session->s_cap_gen);
1142 di->lease_seq = le32_to_cpu(lease->seq);
1143 di->lease_renew_after = half_ttl;
1144 di->lease_renew_from = 0;
1145 di->time = ttl;
1146
1147 __ceph_dentry_lease_touch(di);
1148}
1149
1150static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1151 struct ceph_mds_reply_lease *lease,
1152 struct ceph_mds_session *session,
1153 unsigned long from_time)
1154{
1155 struct ceph_mds_session *old_lease_session = NULL;
1156 spin_lock(&dentry->d_lock);
1157 __update_dentry_lease(dir, dentry, lease, session, from_time,
1158 &old_lease_session);
1159 spin_unlock(&dentry->d_lock);
1160 ceph_put_mds_session(old_lease_session);
1161}
1162
1163/*
1164 * update dentry lease without having parent inode locked
1165 */
1166static void update_dentry_lease_careful(struct dentry *dentry,
1167 struct ceph_mds_reply_lease *lease,
1168 struct ceph_mds_session *session,
1169 unsigned long from_time,
1170 char *dname, u32 dname_len,
1171 struct ceph_vino *pdvino,
1172 struct ceph_vino *ptvino)
1173
1174{
1175 struct inode *dir;
1176 struct ceph_mds_session *old_lease_session = NULL;
1177
1178 spin_lock(&dentry->d_lock);
1179 /* make sure dentry's name matches target */
1180 if (dentry->d_name.len != dname_len ||
1181 memcmp(dentry->d_name.name, dname, dname_len))
1182 goto out_unlock;
1183
1184 dir = d_inode(dentry->d_parent);
1185 /* make sure parent matches dvino */
1186 if (!ceph_ino_compare(dir, pdvino))
1187 goto out_unlock;
1188
1189 /* make sure dentry's inode matches target. NULL ptvino means that
1190 * we expect a negative dentry */
1191 if (ptvino) {
1192 if (d_really_is_negative(dentry))
1193 goto out_unlock;
1194 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1195 goto out_unlock;
1196 } else {
1197 if (d_really_is_positive(dentry))
1198 goto out_unlock;
1199 }
1200
1201 __update_dentry_lease(dir, dentry, lease, session,
1202 from_time, &old_lease_session);
1203out_unlock:
1204 spin_unlock(&dentry->d_lock);
1205 ceph_put_mds_session(old_lease_session);
1206}
1207
1208/*
1209 * splice a dentry to an inode.
1210 * caller must hold directory i_rwsem for this to be safe.
1211 */
1212static int splice_dentry(struct dentry **pdn, struct inode *in)
1213{
1214 struct dentry *dn = *pdn;
1215 struct dentry *realdn;
1216
1217 BUG_ON(d_inode(dn));
1218
1219 if (S_ISDIR(in->i_mode)) {
1220 /* If inode is directory, d_splice_alias() below will remove
1221 * 'realdn' from its origin parent. We need to ensure that
1222 * origin parent's readdir cache will not reference 'realdn'
1223 */
1224 realdn = d_find_any_alias(in);
1225 if (realdn) {
1226 struct ceph_dentry_info *di = ceph_dentry(realdn);
1227 spin_lock(&realdn->d_lock);
1228
1229 realdn->d_op->d_prune(realdn);
1230
1231 di->time = jiffies;
1232 di->lease_shared_gen = 0;
1233 di->offset = 0;
1234
1235 spin_unlock(&realdn->d_lock);
1236 dput(realdn);
1237 }
1238 }
1239
1240 /* dn must be unhashed */
1241 if (!d_unhashed(dn))
1242 d_drop(dn);
1243 realdn = d_splice_alias(in, dn);
1244 if (IS_ERR(realdn)) {
1245 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1246 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1247 return PTR_ERR(realdn);
1248 }
1249
1250 if (realdn) {
1251 dout("dn %p (%d) spliced with %p (%d) "
1252 "inode %p ino %llx.%llx\n",
1253 dn, d_count(dn),
1254 realdn, d_count(realdn),
1255 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1256 dput(dn);
1257 *pdn = realdn;
1258 } else {
1259 BUG_ON(!ceph_dentry(dn));
1260 dout("dn %p attached to %p ino %llx.%llx\n",
1261 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1262 }
1263 return 0;
1264}
1265
1266/*
1267 * Incorporate results into the local cache. This is either just
1268 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1269 * after a lookup).
1270 *
1271 * A reply may contain
1272 * a directory inode along with a dentry.
1273 * and/or a target inode
1274 *
1275 * Called with snap_rwsem (read).
1276 */
1277int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1278{
1279 struct ceph_mds_session *session = req->r_session;
1280 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1281 struct inode *in = NULL;
1282 struct ceph_vino tvino, dvino;
1283 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1284 int err = 0;
1285
1286 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1287 rinfo->head->is_dentry, rinfo->head->is_target);
1288
1289 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1290 dout("fill_trace reply is empty!\n");
1291 if (rinfo->head->result == 0 && req->r_parent)
1292 ceph_invalidate_dir_request(req);
1293 return 0;
1294 }
1295
1296 if (rinfo->head->is_dentry) {
1297 struct inode *dir = req->r_parent;
1298
1299 if (dir) {
1300 err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1301 rinfo->dirfrag, session, -1,
1302 &req->r_caps_reservation);
1303 if (err < 0)
1304 goto done;
1305 } else {
1306 WARN_ON_ONCE(1);
1307 }
1308
1309 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1310 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1311 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1312 struct qstr dname;
1313 struct dentry *dn, *parent;
1314
1315 BUG_ON(!rinfo->head->is_target);
1316 BUG_ON(req->r_dentry);
1317
1318 parent = d_find_any_alias(dir);
1319 BUG_ON(!parent);
1320
1321 dname.name = rinfo->dname;
1322 dname.len = rinfo->dname_len;
1323 dname.hash = full_name_hash(parent, dname.name, dname.len);
1324 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1325 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1326retry_lookup:
1327 dn = d_lookup(parent, &dname);
1328 dout("d_lookup on parent=%p name=%.*s got %p\n",
1329 parent, dname.len, dname.name, dn);
1330
1331 if (!dn) {
1332 dn = d_alloc(parent, &dname);
1333 dout("d_alloc %p '%.*s' = %p\n", parent,
1334 dname.len, dname.name, dn);
1335 if (!dn) {
1336 dput(parent);
1337 err = -ENOMEM;
1338 goto done;
1339 }
1340 err = 0;
1341 } else if (d_really_is_positive(dn) &&
1342 (ceph_ino(d_inode(dn)) != tvino.ino ||
1343 ceph_snap(d_inode(dn)) != tvino.snap)) {
1344 dout(" dn %p points to wrong inode %p\n",
1345 dn, d_inode(dn));
1346 ceph_dir_clear_ordered(dir);
1347 d_delete(dn);
1348 dput(dn);
1349 goto retry_lookup;
1350 }
1351
1352 req->r_dentry = dn;
1353 dput(parent);
1354 }
1355 }
1356
1357 if (rinfo->head->is_target) {
1358 /* Should be filled in by handle_reply */
1359 BUG_ON(!req->r_target_inode);
1360
1361 in = req->r_target_inode;
1362 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1363 NULL, session,
1364 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1365 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1366 rinfo->head->result == 0) ? req->r_fmode : -1,
1367 &req->r_caps_reservation);
1368 if (err < 0) {
1369 pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1370 in, ceph_vinop(in));
1371 req->r_target_inode = NULL;
1372 if (in->i_state & I_NEW)
1373 discard_new_inode(in);
1374 else
1375 iput(in);
1376 goto done;
1377 }
1378 if (in->i_state & I_NEW)
1379 unlock_new_inode(in);
1380 }
1381
1382 /*
1383 * ignore null lease/binding on snapdir ENOENT, or else we
1384 * will have trouble splicing in the virtual snapdir later
1385 */
1386 if (rinfo->head->is_dentry &&
1387 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1388 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1389 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1390 fsc->mount_options->snapdir_name,
1391 req->r_dentry->d_name.len))) {
1392 /*
1393 * lookup link rename : null -> possibly existing inode
1394 * mknod symlink mkdir : null -> new inode
1395 * unlink : linked -> null
1396 */
1397 struct inode *dir = req->r_parent;
1398 struct dentry *dn = req->r_dentry;
1399 bool have_dir_cap, have_lease;
1400
1401 BUG_ON(!dn);
1402 BUG_ON(!dir);
1403 BUG_ON(d_inode(dn->d_parent) != dir);
1404
1405 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1406 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1407
1408 BUG_ON(ceph_ino(dir) != dvino.ino);
1409 BUG_ON(ceph_snap(dir) != dvino.snap);
1410
1411 /* do we have a lease on the whole dir? */
1412 have_dir_cap =
1413 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1414 CEPH_CAP_FILE_SHARED);
1415
1416 /* do we have a dn lease? */
1417 have_lease = have_dir_cap ||
1418 le32_to_cpu(rinfo->dlease->duration_ms);
1419 if (!have_lease)
1420 dout("fill_trace no dentry lease or dir cap\n");
1421
1422 /* rename? */
1423 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1424 struct inode *olddir = req->r_old_dentry_dir;
1425 BUG_ON(!olddir);
1426
1427 dout(" src %p '%pd' dst %p '%pd'\n",
1428 req->r_old_dentry,
1429 req->r_old_dentry,
1430 dn, dn);
1431 dout("fill_trace doing d_move %p -> %p\n",
1432 req->r_old_dentry, dn);
1433
1434 /* d_move screws up sibling dentries' offsets */
1435 ceph_dir_clear_ordered(dir);
1436 ceph_dir_clear_ordered(olddir);
1437
1438 d_move(req->r_old_dentry, dn);
1439 dout(" src %p '%pd' dst %p '%pd'\n",
1440 req->r_old_dentry,
1441 req->r_old_dentry,
1442 dn, dn);
1443
1444 /* ensure target dentry is invalidated, despite
1445 rehashing bug in vfs_rename_dir */
1446 ceph_invalidate_dentry_lease(dn);
1447
1448 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1449 ceph_dentry(req->r_old_dentry)->offset);
1450
1451 /* swap r_dentry and r_old_dentry in case that
1452 * splice_dentry() gets called later. This is safe
1453 * because no other place will use them */
1454 req->r_dentry = req->r_old_dentry;
1455 req->r_old_dentry = dn;
1456 dn = req->r_dentry;
1457 }
1458
1459 /* null dentry? */
1460 if (!rinfo->head->is_target) {
1461 dout("fill_trace null dentry\n");
1462 if (d_really_is_positive(dn)) {
1463 dout("d_delete %p\n", dn);
1464 ceph_dir_clear_ordered(dir);
1465 d_delete(dn);
1466 } else if (have_lease) {
1467 if (d_unhashed(dn))
1468 d_add(dn, NULL);
1469 }
1470
1471 if (!d_unhashed(dn) && have_lease)
1472 update_dentry_lease(dir, dn,
1473 rinfo->dlease, session,
1474 req->r_request_started);
1475 goto done;
1476 }
1477
1478 /* attach proper inode */
1479 if (d_really_is_negative(dn)) {
1480 ceph_dir_clear_ordered(dir);
1481 ihold(in);
1482 err = splice_dentry(&req->r_dentry, in);
1483 if (err < 0)
1484 goto done;
1485 dn = req->r_dentry; /* may have spliced */
1486 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1487 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1488 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1489 ceph_vinop(in));
1490 d_invalidate(dn);
1491 have_lease = false;
1492 }
1493
1494 if (have_lease) {
1495 update_dentry_lease(dir, dn,
1496 rinfo->dlease, session,
1497 req->r_request_started);
1498 }
1499 dout(" final dn %p\n", dn);
1500 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1501 req->r_op == CEPH_MDS_OP_MKSNAP) &&
1502 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1503 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1504 struct inode *dir = req->r_parent;
1505
1506 /* fill out a snapdir LOOKUPSNAP dentry */
1507 BUG_ON(!dir);
1508 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1509 BUG_ON(!req->r_dentry);
1510 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1511 ceph_dir_clear_ordered(dir);
1512 ihold(in);
1513 err = splice_dentry(&req->r_dentry, in);
1514 if (err < 0)
1515 goto done;
1516 } else if (rinfo->head->is_dentry && req->r_dentry) {
1517 /* parent inode is not locked, be carefull */
1518 struct ceph_vino *ptvino = NULL;
1519 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1520 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1521 if (rinfo->head->is_target) {
1522 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1523 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1524 ptvino = &tvino;
1525 }
1526 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1527 session, req->r_request_started,
1528 rinfo->dname, rinfo->dname_len,
1529 &dvino, ptvino);
1530 }
1531done:
1532 dout("fill_trace done err=%d\n", err);
1533 return err;
1534}
1535
1536/*
1537 * Prepopulate our cache with readdir results, leases, etc.
1538 */
1539static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1540 struct ceph_mds_session *session)
1541{
1542 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1543 int i, err = 0;
1544
1545 for (i = 0; i < rinfo->dir_nr; i++) {
1546 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1547 struct ceph_vino vino;
1548 struct inode *in;
1549 int rc;
1550
1551 vino.ino = le64_to_cpu(rde->inode.in->ino);
1552 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1553
1554 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1555 if (IS_ERR(in)) {
1556 err = PTR_ERR(in);
1557 dout("new_inode badness got %d\n", err);
1558 continue;
1559 }
1560 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1561 -1, &req->r_caps_reservation);
1562 if (rc < 0) {
1563 pr_err("ceph_fill_inode badness on %p got %d\n",
1564 in, rc);
1565 err = rc;
1566 if (in->i_state & I_NEW) {
1567 ihold(in);
1568 discard_new_inode(in);
1569 }
1570 } else if (in->i_state & I_NEW) {
1571 unlock_new_inode(in);
1572 }
1573
1574 iput(in);
1575 }
1576
1577 return err;
1578}
1579
1580void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1581{
1582 if (ctl->page) {
1583 kunmap(ctl->page);
1584 put_page(ctl->page);
1585 ctl->page = NULL;
1586 }
1587}
1588
1589static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1590 struct ceph_readdir_cache_control *ctl,
1591 struct ceph_mds_request *req)
1592{
1593 struct ceph_inode_info *ci = ceph_inode(dir);
1594 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1595 unsigned idx = ctl->index % nsize;
1596 pgoff_t pgoff = ctl->index / nsize;
1597
1598 if (!ctl->page || pgoff != page_index(ctl->page)) {
1599 ceph_readdir_cache_release(ctl);
1600 if (idx == 0)
1601 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1602 else
1603 ctl->page = find_lock_page(&dir->i_data, pgoff);
1604 if (!ctl->page) {
1605 ctl->index = -1;
1606 return idx == 0 ? -ENOMEM : 0;
1607 }
1608 /* reading/filling the cache are serialized by
1609 * i_rwsem, no need to use page lock */
1610 unlock_page(ctl->page);
1611 ctl->dentries = kmap(ctl->page);
1612 if (idx == 0)
1613 memset(ctl->dentries, 0, PAGE_SIZE);
1614 }
1615
1616 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1617 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1618 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1619 ctl->dentries[idx] = dn;
1620 ctl->index++;
1621 } else {
1622 dout("disable readdir cache\n");
1623 ctl->index = -1;
1624 }
1625 return 0;
1626}
1627
1628int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1629 struct ceph_mds_session *session)
1630{
1631 struct dentry *parent = req->r_dentry;
1632 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1633 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1634 struct qstr dname;
1635 struct dentry *dn;
1636 struct inode *in;
1637 int err = 0, skipped = 0, ret, i;
1638 u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1639 u32 last_hash = 0;
1640 u32 fpos_offset;
1641 struct ceph_readdir_cache_control cache_ctl = {};
1642
1643 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1644 return readdir_prepopulate_inodes_only(req, session);
1645
1646 if (rinfo->hash_order) {
1647 if (req->r_path2) {
1648 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1649 req->r_path2,
1650 strlen(req->r_path2));
1651 last_hash = ceph_frag_value(last_hash);
1652 } else if (rinfo->offset_hash) {
1653 /* mds understands offset_hash */
1654 WARN_ON_ONCE(req->r_readdir_offset != 2);
1655 last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1656 }
1657 }
1658
1659 if (rinfo->dir_dir &&
1660 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1661 dout("readdir_prepopulate got new frag %x -> %x\n",
1662 frag, le32_to_cpu(rinfo->dir_dir->frag));
1663 frag = le32_to_cpu(rinfo->dir_dir->frag);
1664 if (!rinfo->hash_order)
1665 req->r_readdir_offset = 2;
1666 }
1667
1668 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1669 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1670 rinfo->dir_nr, parent);
1671 } else {
1672 dout("readdir_prepopulate %d items under dn %p\n",
1673 rinfo->dir_nr, parent);
1674 if (rinfo->dir_dir)
1675 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1676
1677 if (ceph_frag_is_leftmost(frag) &&
1678 req->r_readdir_offset == 2 &&
1679 !(rinfo->hash_order && last_hash)) {
1680 /* note dir version at start of readdir so we can
1681 * tell if any dentries get dropped */
1682 req->r_dir_release_cnt =
1683 atomic64_read(&ci->i_release_count);
1684 req->r_dir_ordered_cnt =
1685 atomic64_read(&ci->i_ordered_count);
1686 req->r_readdir_cache_idx = 0;
1687 }
1688 }
1689
1690 cache_ctl.index = req->r_readdir_cache_idx;
1691 fpos_offset = req->r_readdir_offset;
1692
1693 /* FIXME: release caps/leases if error occurs */
1694 for (i = 0; i < rinfo->dir_nr; i++) {
1695 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1696 struct ceph_vino tvino;
1697
1698 dname.name = rde->name;
1699 dname.len = rde->name_len;
1700 dname.hash = full_name_hash(parent, dname.name, dname.len);
1701
1702 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1703 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1704
1705 if (rinfo->hash_order) {
1706 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1707 rde->name, rde->name_len);
1708 hash = ceph_frag_value(hash);
1709 if (hash != last_hash)
1710 fpos_offset = 2;
1711 last_hash = hash;
1712 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1713 } else {
1714 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1715 }
1716
1717retry_lookup:
1718 dn = d_lookup(parent, &dname);
1719 dout("d_lookup on parent=%p name=%.*s got %p\n",
1720 parent, dname.len, dname.name, dn);
1721
1722 if (!dn) {
1723 dn = d_alloc(parent, &dname);
1724 dout("d_alloc %p '%.*s' = %p\n", parent,
1725 dname.len, dname.name, dn);
1726 if (!dn) {
1727 dout("d_alloc badness\n");
1728 err = -ENOMEM;
1729 goto out;
1730 }
1731 } else if (d_really_is_positive(dn) &&
1732 (ceph_ino(d_inode(dn)) != tvino.ino ||
1733 ceph_snap(d_inode(dn)) != tvino.snap)) {
1734 struct ceph_dentry_info *di = ceph_dentry(dn);
1735 dout(" dn %p points to wrong inode %p\n",
1736 dn, d_inode(dn));
1737
1738 spin_lock(&dn->d_lock);
1739 if (di->offset > 0 &&
1740 di->lease_shared_gen ==
1741 atomic_read(&ci->i_shared_gen)) {
1742 __ceph_dir_clear_ordered(ci);
1743 di->offset = 0;
1744 }
1745 spin_unlock(&dn->d_lock);
1746
1747 d_delete(dn);
1748 dput(dn);
1749 goto retry_lookup;
1750 }
1751
1752 /* inode */
1753 if (d_really_is_positive(dn)) {
1754 in = d_inode(dn);
1755 } else {
1756 in = ceph_get_inode(parent->d_sb, tvino);
1757 if (IS_ERR(in)) {
1758 dout("new_inode badness\n");
1759 d_drop(dn);
1760 dput(dn);
1761 err = PTR_ERR(in);
1762 goto out;
1763 }
1764 }
1765
1766 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1767 -1, &req->r_caps_reservation);
1768 if (ret < 0) {
1769 pr_err("ceph_fill_inode badness on %p\n", in);
1770 if (d_really_is_negative(dn)) {
1771 if (in->i_state & I_NEW) {
1772 ihold(in);
1773 discard_new_inode(in);
1774 }
1775 iput(in);
1776 }
1777 d_drop(dn);
1778 err = ret;
1779 goto next_item;
1780 }
1781 if (in->i_state & I_NEW)
1782 unlock_new_inode(in);
1783
1784 if (d_really_is_negative(dn)) {
1785 if (ceph_security_xattr_deadlock(in)) {
1786 dout(" skip splicing dn %p to inode %p"
1787 " (security xattr deadlock)\n", dn, in);
1788 iput(in);
1789 skipped++;
1790 goto next_item;
1791 }
1792
1793 err = splice_dentry(&dn, in);
1794 if (err < 0)
1795 goto next_item;
1796 }
1797
1798 ceph_dentry(dn)->offset = rde->offset;
1799
1800 update_dentry_lease(d_inode(parent), dn,
1801 rde->lease, req->r_session,
1802 req->r_request_started);
1803
1804 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1805 ret = fill_readdir_cache(d_inode(parent), dn,
1806 &cache_ctl, req);
1807 if (ret < 0)
1808 err = ret;
1809 }
1810next_item:
1811 dput(dn);
1812 }
1813out:
1814 if (err == 0 && skipped == 0) {
1815 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1816 req->r_readdir_cache_idx = cache_ctl.index;
1817 }
1818 ceph_readdir_cache_release(&cache_ctl);
1819 dout("readdir_prepopulate done\n");
1820 return err;
1821}
1822
1823bool ceph_inode_set_size(struct inode *inode, loff_t size)
1824{
1825 struct ceph_inode_info *ci = ceph_inode(inode);
1826 bool ret;
1827
1828 spin_lock(&ci->i_ceph_lock);
1829 dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
1830 i_size_write(inode, size);
1831 ceph_fscache_update(inode);
1832 inode->i_blocks = calc_inode_blocks(size);
1833
1834 ret = __ceph_should_report_size(ci);
1835
1836 spin_unlock(&ci->i_ceph_lock);
1837
1838 return ret;
1839}
1840
1841void ceph_queue_inode_work(struct inode *inode, int work_bit)
1842{
1843 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1844 struct ceph_inode_info *ci = ceph_inode(inode);
1845 set_bit(work_bit, &ci->i_work_mask);
1846
1847 ihold(inode);
1848 if (queue_work(fsc->inode_wq, &ci->i_work)) {
1849 dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask);
1850 } else {
1851 dout("queue_inode_work %p already queued, mask=%lx\n",
1852 inode, ci->i_work_mask);
1853 iput(inode);
1854 }
1855}
1856
1857static void ceph_do_invalidate_pages(struct inode *inode)
1858{
1859 struct ceph_inode_info *ci = ceph_inode(inode);
1860 u32 orig_gen;
1861 int check = 0;
1862
1863 ceph_fscache_invalidate(inode, false);
1864
1865 mutex_lock(&ci->i_truncate_mutex);
1866
1867 if (ceph_inode_is_shutdown(inode)) {
1868 pr_warn_ratelimited("%s: inode %llx.%llx is shut down\n",
1869 __func__, ceph_vinop(inode));
1870 mapping_set_error(inode->i_mapping, -EIO);
1871 truncate_pagecache(inode, 0);
1872 mutex_unlock(&ci->i_truncate_mutex);
1873 goto out;
1874 }
1875
1876 spin_lock(&ci->i_ceph_lock);
1877 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1878 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1879 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1880 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1881 check = 1;
1882 spin_unlock(&ci->i_ceph_lock);
1883 mutex_unlock(&ci->i_truncate_mutex);
1884 goto out;
1885 }
1886 orig_gen = ci->i_rdcache_gen;
1887 spin_unlock(&ci->i_ceph_lock);
1888
1889 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1890 pr_err("invalidate_inode_pages2 %llx.%llx failed\n",
1891 ceph_vinop(inode));
1892 }
1893
1894 spin_lock(&ci->i_ceph_lock);
1895 if (orig_gen == ci->i_rdcache_gen &&
1896 orig_gen == ci->i_rdcache_revoking) {
1897 dout("invalidate_pages %p gen %d successful\n", inode,
1898 ci->i_rdcache_gen);
1899 ci->i_rdcache_revoking--;
1900 check = 1;
1901 } else {
1902 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1903 inode, orig_gen, ci->i_rdcache_gen,
1904 ci->i_rdcache_revoking);
1905 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1906 check = 1;
1907 }
1908 spin_unlock(&ci->i_ceph_lock);
1909 mutex_unlock(&ci->i_truncate_mutex);
1910out:
1911 if (check)
1912 ceph_check_caps(ci, 0);
1913}
1914
1915/*
1916 * Make sure any pending truncation is applied before doing anything
1917 * that may depend on it.
1918 */
1919void __ceph_do_pending_vmtruncate(struct inode *inode)
1920{
1921 struct ceph_inode_info *ci = ceph_inode(inode);
1922 u64 to;
1923 int wrbuffer_refs, finish = 0;
1924
1925 mutex_lock(&ci->i_truncate_mutex);
1926retry:
1927 spin_lock(&ci->i_ceph_lock);
1928 if (ci->i_truncate_pending == 0) {
1929 dout("__do_pending_vmtruncate %p none pending\n", inode);
1930 spin_unlock(&ci->i_ceph_lock);
1931 mutex_unlock(&ci->i_truncate_mutex);
1932 return;
1933 }
1934
1935 /*
1936 * make sure any dirty snapped pages are flushed before we
1937 * possibly truncate them.. so write AND block!
1938 */
1939 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1940 spin_unlock(&ci->i_ceph_lock);
1941 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1942 inode);
1943 filemap_write_and_wait_range(&inode->i_data, 0,
1944 inode->i_sb->s_maxbytes);
1945 goto retry;
1946 }
1947
1948 /* there should be no reader or writer */
1949 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1950
1951 to = ci->i_truncate_size;
1952 wrbuffer_refs = ci->i_wrbuffer_ref;
1953 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1954 ci->i_truncate_pending, to);
1955 spin_unlock(&ci->i_ceph_lock);
1956
1957 ceph_fscache_resize(inode, to);
1958 truncate_pagecache(inode, to);
1959
1960 spin_lock(&ci->i_ceph_lock);
1961 if (to == ci->i_truncate_size) {
1962 ci->i_truncate_pending = 0;
1963 finish = 1;
1964 }
1965 spin_unlock(&ci->i_ceph_lock);
1966 if (!finish)
1967 goto retry;
1968
1969 mutex_unlock(&ci->i_truncate_mutex);
1970
1971 if (wrbuffer_refs == 0)
1972 ceph_check_caps(ci, 0);
1973
1974 wake_up_all(&ci->i_cap_wq);
1975}
1976
1977static void ceph_inode_work(struct work_struct *work)
1978{
1979 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1980 i_work);
1981 struct inode *inode = &ci->netfs.inode;
1982
1983 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1984 dout("writeback %p\n", inode);
1985 filemap_fdatawrite(&inode->i_data);
1986 }
1987 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1988 ceph_do_invalidate_pages(inode);
1989
1990 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1991 __ceph_do_pending_vmtruncate(inode);
1992
1993 if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
1994 ceph_check_caps(ci, 0);
1995
1996 if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
1997 ceph_flush_snaps(ci, NULL);
1998
1999 iput(inode);
2000}
2001
2002/*
2003 * symlinks
2004 */
2005static const struct inode_operations ceph_symlink_iops = {
2006 .get_link = simple_get_link,
2007 .setattr = ceph_setattr,
2008 .getattr = ceph_getattr,
2009 .listxattr = ceph_listxattr,
2010};
2011
2012int __ceph_setattr(struct inode *inode, struct iattr *attr)
2013{
2014 struct ceph_inode_info *ci = ceph_inode(inode);
2015 unsigned int ia_valid = attr->ia_valid;
2016 struct ceph_mds_request *req;
2017 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2018 struct ceph_cap_flush *prealloc_cf;
2019 int issued;
2020 int release = 0, dirtied = 0;
2021 int mask = 0;
2022 int err = 0;
2023 int inode_dirty_flags = 0;
2024 bool lock_snap_rwsem = false;
2025
2026 prealloc_cf = ceph_alloc_cap_flush();
2027 if (!prealloc_cf)
2028 return -ENOMEM;
2029
2030 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2031 USE_AUTH_MDS);
2032 if (IS_ERR(req)) {
2033 ceph_free_cap_flush(prealloc_cf);
2034 return PTR_ERR(req);
2035 }
2036
2037 spin_lock(&ci->i_ceph_lock);
2038 issued = __ceph_caps_issued(ci, NULL);
2039
2040 if (!ci->i_head_snapc &&
2041 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2042 lock_snap_rwsem = true;
2043 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2044 spin_unlock(&ci->i_ceph_lock);
2045 down_read(&mdsc->snap_rwsem);
2046 spin_lock(&ci->i_ceph_lock);
2047 issued = __ceph_caps_issued(ci, NULL);
2048 }
2049 }
2050
2051 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2052
2053 if (ia_valid & ATTR_UID) {
2054 dout("setattr %p uid %d -> %d\n", inode,
2055 from_kuid(&init_user_ns, inode->i_uid),
2056 from_kuid(&init_user_ns, attr->ia_uid));
2057 if (issued & CEPH_CAP_AUTH_EXCL) {
2058 inode->i_uid = attr->ia_uid;
2059 dirtied |= CEPH_CAP_AUTH_EXCL;
2060 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2061 !uid_eq(attr->ia_uid, inode->i_uid)) {
2062 req->r_args.setattr.uid = cpu_to_le32(
2063 from_kuid(&init_user_ns, attr->ia_uid));
2064 mask |= CEPH_SETATTR_UID;
2065 release |= CEPH_CAP_AUTH_SHARED;
2066 }
2067 }
2068 if (ia_valid & ATTR_GID) {
2069 dout("setattr %p gid %d -> %d\n", inode,
2070 from_kgid(&init_user_ns, inode->i_gid),
2071 from_kgid(&init_user_ns, attr->ia_gid));
2072 if (issued & CEPH_CAP_AUTH_EXCL) {
2073 inode->i_gid = attr->ia_gid;
2074 dirtied |= CEPH_CAP_AUTH_EXCL;
2075 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2076 !gid_eq(attr->ia_gid, inode->i_gid)) {
2077 req->r_args.setattr.gid = cpu_to_le32(
2078 from_kgid(&init_user_ns, attr->ia_gid));
2079 mask |= CEPH_SETATTR_GID;
2080 release |= CEPH_CAP_AUTH_SHARED;
2081 }
2082 }
2083 if (ia_valid & ATTR_MODE) {
2084 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2085 attr->ia_mode);
2086 if (issued & CEPH_CAP_AUTH_EXCL) {
2087 inode->i_mode = attr->ia_mode;
2088 dirtied |= CEPH_CAP_AUTH_EXCL;
2089 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2090 attr->ia_mode != inode->i_mode) {
2091 inode->i_mode = attr->ia_mode;
2092 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2093 mask |= CEPH_SETATTR_MODE;
2094 release |= CEPH_CAP_AUTH_SHARED;
2095 }
2096 }
2097
2098 if (ia_valid & ATTR_ATIME) {
2099 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2100 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2101 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2102 if (issued & CEPH_CAP_FILE_EXCL) {
2103 ci->i_time_warp_seq++;
2104 inode->i_atime = attr->ia_atime;
2105 dirtied |= CEPH_CAP_FILE_EXCL;
2106 } else if ((issued & CEPH_CAP_FILE_WR) &&
2107 timespec64_compare(&inode->i_atime,
2108 &attr->ia_atime) < 0) {
2109 inode->i_atime = attr->ia_atime;
2110 dirtied |= CEPH_CAP_FILE_WR;
2111 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2112 !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2113 ceph_encode_timespec64(&req->r_args.setattr.atime,
2114 &attr->ia_atime);
2115 mask |= CEPH_SETATTR_ATIME;
2116 release |= CEPH_CAP_FILE_SHARED |
2117 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2118 }
2119 }
2120 if (ia_valid & ATTR_SIZE) {
2121 loff_t isize = i_size_read(inode);
2122
2123 dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
2124 if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
2125 if (attr->ia_size > isize) {
2126 i_size_write(inode, attr->ia_size);
2127 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2128 ci->i_reported_size = attr->ia_size;
2129 dirtied |= CEPH_CAP_FILE_EXCL;
2130 ia_valid |= ATTR_MTIME;
2131 }
2132 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2133 attr->ia_size != isize) {
2134 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2135 req->r_args.setattr.old_size = cpu_to_le64(isize);
2136 mask |= CEPH_SETATTR_SIZE;
2137 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2138 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2139 }
2140 }
2141 if (ia_valid & ATTR_MTIME) {
2142 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2143 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2144 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2145 if (issued & CEPH_CAP_FILE_EXCL) {
2146 ci->i_time_warp_seq++;
2147 inode->i_mtime = attr->ia_mtime;
2148 dirtied |= CEPH_CAP_FILE_EXCL;
2149 } else if ((issued & CEPH_CAP_FILE_WR) &&
2150 timespec64_compare(&inode->i_mtime,
2151 &attr->ia_mtime) < 0) {
2152 inode->i_mtime = attr->ia_mtime;
2153 dirtied |= CEPH_CAP_FILE_WR;
2154 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2155 !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2156 ceph_encode_timespec64(&req->r_args.setattr.mtime,
2157 &attr->ia_mtime);
2158 mask |= CEPH_SETATTR_MTIME;
2159 release |= CEPH_CAP_FILE_SHARED |
2160 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2161 }
2162 }
2163
2164 /* these do nothing */
2165 if (ia_valid & ATTR_CTIME) {
2166 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2167 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2168 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2169 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2170 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2171 only ? "ctime only" : "ignored");
2172 if (only) {
2173 /*
2174 * if kernel wants to dirty ctime but nothing else,
2175 * we need to choose a cap to dirty under, or do
2176 * a almost-no-op setattr
2177 */
2178 if (issued & CEPH_CAP_AUTH_EXCL)
2179 dirtied |= CEPH_CAP_AUTH_EXCL;
2180 else if (issued & CEPH_CAP_FILE_EXCL)
2181 dirtied |= CEPH_CAP_FILE_EXCL;
2182 else if (issued & CEPH_CAP_XATTR_EXCL)
2183 dirtied |= CEPH_CAP_XATTR_EXCL;
2184 else
2185 mask |= CEPH_SETATTR_CTIME;
2186 }
2187 }
2188 if (ia_valid & ATTR_FILE)
2189 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2190
2191 if (dirtied) {
2192 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2193 &prealloc_cf);
2194 inode->i_ctime = attr->ia_ctime;
2195 inode_inc_iversion_raw(inode);
2196 }
2197
2198 release &= issued;
2199 spin_unlock(&ci->i_ceph_lock);
2200 if (lock_snap_rwsem)
2201 up_read(&mdsc->snap_rwsem);
2202
2203 if (inode_dirty_flags)
2204 __mark_inode_dirty(inode, inode_dirty_flags);
2205
2206 if (mask) {
2207 req->r_inode = inode;
2208 ihold(inode);
2209 req->r_inode_drop = release;
2210 req->r_args.setattr.mask = cpu_to_le32(mask);
2211 req->r_num_caps = 1;
2212 req->r_stamp = attr->ia_ctime;
2213 err = ceph_mdsc_do_request(mdsc, NULL, req);
2214 }
2215 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2216 ceph_cap_string(dirtied), mask);
2217
2218 ceph_mdsc_put_request(req);
2219 ceph_free_cap_flush(prealloc_cf);
2220
2221 if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2222 __ceph_do_pending_vmtruncate(inode);
2223
2224 return err;
2225}
2226
2227/*
2228 * setattr
2229 */
2230int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
2231 struct iattr *attr)
2232{
2233 struct inode *inode = d_inode(dentry);
2234 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2235 int err;
2236
2237 if (ceph_snap(inode) != CEPH_NOSNAP)
2238 return -EROFS;
2239
2240 if (ceph_inode_is_shutdown(inode))
2241 return -ESTALE;
2242
2243 err = setattr_prepare(&init_user_ns, dentry, attr);
2244 if (err != 0)
2245 return err;
2246
2247 if ((attr->ia_valid & ATTR_SIZE) &&
2248 attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2249 return -EFBIG;
2250
2251 if ((attr->ia_valid & ATTR_SIZE) &&
2252 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2253 return -EDQUOT;
2254
2255 err = __ceph_setattr(inode, attr);
2256
2257 if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2258 err = posix_acl_chmod(&init_user_ns, dentry, attr->ia_mode);
2259
2260 return err;
2261}
2262
2263int ceph_try_to_choose_auth_mds(struct inode *inode, int mask)
2264{
2265 int issued = ceph_caps_issued(ceph_inode(inode));
2266
2267 /*
2268 * If any 'x' caps is issued we can just choose the auth MDS
2269 * instead of the random replica MDSes. Because only when the
2270 * Locker is in LOCK_EXEC state will the loner client could
2271 * get the 'x' caps. And if we send the getattr requests to
2272 * any replica MDS it must auth pin and tries to rdlock from
2273 * the auth MDS, and then the auth MDS need to do the Locker
2274 * state transition to LOCK_SYNC. And after that the lock state
2275 * will change back.
2276 *
2277 * This cost much when doing the Locker state transition and
2278 * usually will need to revoke caps from clients.
2279 *
2280 * And for the 'Xs' caps for getxattr we will also choose the
2281 * auth MDS, because the MDS side code is buggy due to setxattr
2282 * won't notify the replica MDSes when the values changed and
2283 * the replica MDS will return the old values. Though we will
2284 * fix it in MDS code, but this still makes sense for old ceph.
2285 */
2286 if (((mask & CEPH_CAP_ANY_SHARED) && (issued & CEPH_CAP_ANY_EXCL))
2287 || (mask & (CEPH_STAT_RSTAT | CEPH_STAT_CAP_XATTR)))
2288 return USE_AUTH_MDS;
2289 else
2290 return USE_ANY_MDS;
2291}
2292
2293/*
2294 * Verify that we have a lease on the given mask. If not,
2295 * do a getattr against an mds.
2296 */
2297int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2298 int mask, bool force)
2299{
2300 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2301 struct ceph_mds_client *mdsc = fsc->mdsc;
2302 struct ceph_mds_request *req;
2303 int mode;
2304 int err;
2305
2306 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2307 dout("do_getattr inode %p SNAPDIR\n", inode);
2308 return 0;
2309 }
2310
2311 dout("do_getattr inode %p mask %s mode 0%o\n",
2312 inode, ceph_cap_string(mask), inode->i_mode);
2313 if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2314 return 0;
2315
2316 mode = ceph_try_to_choose_auth_mds(inode, mask);
2317 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2318 if (IS_ERR(req))
2319 return PTR_ERR(req);
2320 req->r_inode = inode;
2321 ihold(inode);
2322 req->r_num_caps = 1;
2323 req->r_args.getattr.mask = cpu_to_le32(mask);
2324 req->r_locked_page = locked_page;
2325 err = ceph_mdsc_do_request(mdsc, NULL, req);
2326 if (locked_page && err == 0) {
2327 u64 inline_version = req->r_reply_info.targeti.inline_version;
2328 if (inline_version == 0) {
2329 /* the reply is supposed to contain inline data */
2330 err = -EINVAL;
2331 } else if (inline_version == CEPH_INLINE_NONE ||
2332 inline_version == 1) {
2333 err = -ENODATA;
2334 } else {
2335 err = req->r_reply_info.targeti.inline_len;
2336 }
2337 }
2338 ceph_mdsc_put_request(req);
2339 dout("do_getattr result=%d\n", err);
2340 return err;
2341}
2342
2343int ceph_do_getvxattr(struct inode *inode, const char *name, void *value,
2344 size_t size)
2345{
2346 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2347 struct ceph_mds_client *mdsc = fsc->mdsc;
2348 struct ceph_mds_request *req;
2349 int mode = USE_AUTH_MDS;
2350 int err;
2351 char *xattr_value;
2352 size_t xattr_value_len;
2353
2354 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETVXATTR, mode);
2355 if (IS_ERR(req)) {
2356 err = -ENOMEM;
2357 goto out;
2358 }
2359
2360 req->r_feature_needed = CEPHFS_FEATURE_OP_GETVXATTR;
2361 req->r_path2 = kstrdup(name, GFP_NOFS);
2362 if (!req->r_path2) {
2363 err = -ENOMEM;
2364 goto put;
2365 }
2366
2367 ihold(inode);
2368 req->r_inode = inode;
2369 err = ceph_mdsc_do_request(mdsc, NULL, req);
2370 if (err < 0)
2371 goto put;
2372
2373 xattr_value = req->r_reply_info.xattr_info.xattr_value;
2374 xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len;
2375
2376 dout("do_getvxattr xattr_value_len:%zu, size:%zu\n", xattr_value_len, size);
2377
2378 err = (int)xattr_value_len;
2379 if (size == 0)
2380 goto put;
2381
2382 if (xattr_value_len > size) {
2383 err = -ERANGE;
2384 goto put;
2385 }
2386
2387 memcpy(value, xattr_value, xattr_value_len);
2388put:
2389 ceph_mdsc_put_request(req);
2390out:
2391 dout("do_getvxattr result=%d\n", err);
2392 return err;
2393}
2394
2395
2396/*
2397 * Check inode permissions. We verify we have a valid value for
2398 * the AUTH cap, then call the generic handler.
2399 */
2400int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode,
2401 int mask)
2402{
2403 int err;
2404
2405 if (mask & MAY_NOT_BLOCK)
2406 return -ECHILD;
2407
2408 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2409
2410 if (!err)
2411 err = generic_permission(&init_user_ns, inode, mask);
2412 return err;
2413}
2414
2415/* Craft a mask of needed caps given a set of requested statx attrs. */
2416static int statx_to_caps(u32 want, umode_t mode)
2417{
2418 int mask = 0;
2419
2420 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2421 mask |= CEPH_CAP_AUTH_SHARED;
2422
2423 if (want & (STATX_NLINK|STATX_CTIME)) {
2424 /*
2425 * The link count for directories depends on inode->i_subdirs,
2426 * and that is only updated when Fs caps are held.
2427 */
2428 if (S_ISDIR(mode))
2429 mask |= CEPH_CAP_FILE_SHARED;
2430 else
2431 mask |= CEPH_CAP_LINK_SHARED;
2432 }
2433
2434 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2435 STATX_BLOCKS))
2436 mask |= CEPH_CAP_FILE_SHARED;
2437
2438 if (want & (STATX_CTIME))
2439 mask |= CEPH_CAP_XATTR_SHARED;
2440
2441 return mask;
2442}
2443
2444/*
2445 * Get all the attributes. If we have sufficient caps for the requested attrs,
2446 * then we can avoid talking to the MDS at all.
2447 */
2448int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path,
2449 struct kstat *stat, u32 request_mask, unsigned int flags)
2450{
2451 struct inode *inode = d_inode(path->dentry);
2452 struct super_block *sb = inode->i_sb;
2453 struct ceph_inode_info *ci = ceph_inode(inode);
2454 u32 valid_mask = STATX_BASIC_STATS;
2455 int err = 0;
2456
2457 if (ceph_inode_is_shutdown(inode))
2458 return -ESTALE;
2459
2460 /* Skip the getattr altogether if we're asked not to sync */
2461 if ((flags & AT_STATX_SYNC_TYPE) != AT_STATX_DONT_SYNC) {
2462 err = ceph_do_getattr(inode,
2463 statx_to_caps(request_mask, inode->i_mode),
2464 flags & AT_STATX_FORCE_SYNC);
2465 if (err)
2466 return err;
2467 }
2468
2469 generic_fillattr(&init_user_ns, inode, stat);
2470 stat->ino = ceph_present_inode(inode);
2471
2472 /*
2473 * btime on newly-allocated inodes is 0, so if this is still set to
2474 * that, then assume that it's not valid.
2475 */
2476 if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2477 stat->btime = ci->i_btime;
2478 valid_mask |= STATX_BTIME;
2479 }
2480
2481 if (ceph_snap(inode) == CEPH_NOSNAP)
2482 stat->dev = sb->s_dev;
2483 else
2484 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2485
2486 if (S_ISDIR(inode->i_mode)) {
2487 if (ceph_test_mount_opt(ceph_sb_to_client(sb), RBYTES)) {
2488 stat->size = ci->i_rbytes;
2489 } else if (ceph_snap(inode) == CEPH_SNAPDIR) {
2490 struct ceph_inode_info *pci;
2491 struct ceph_snap_realm *realm;
2492 struct inode *parent;
2493
2494 parent = ceph_lookup_inode(sb, ceph_ino(inode));
2495 if (IS_ERR(parent))
2496 return PTR_ERR(parent);
2497
2498 pci = ceph_inode(parent);
2499 spin_lock(&pci->i_ceph_lock);
2500 realm = pci->i_snap_realm;
2501 if (realm)
2502 stat->size = realm->num_snaps;
2503 else
2504 stat->size = 0;
2505 spin_unlock(&pci->i_ceph_lock);
2506 iput(parent);
2507 } else {
2508 stat->size = ci->i_files + ci->i_subdirs;
2509 }
2510 stat->blocks = 0;
2511 stat->blksize = 65536;
2512 /*
2513 * Some applications rely on the number of st_nlink
2514 * value on directories to be either 0 (if unlinked)
2515 * or 2 + number of subdirectories.
2516 */
2517 if (stat->nlink == 1)
2518 /* '.' + '..' + subdirs */
2519 stat->nlink = 1 + 1 + ci->i_subdirs;
2520 }
2521
2522 stat->result_mask = request_mask & valid_mask;
2523 return err;
2524}
2525
2526void ceph_inode_shutdown(struct inode *inode)
2527{
2528 struct ceph_inode_info *ci = ceph_inode(inode);
2529 struct rb_node *p;
2530 int iputs = 0;
2531 bool invalidate = false;
2532
2533 spin_lock(&ci->i_ceph_lock);
2534 ci->i_ceph_flags |= CEPH_I_SHUTDOWN;
2535 p = rb_first(&ci->i_caps);
2536 while (p) {
2537 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
2538
2539 p = rb_next(p);
2540 iputs += ceph_purge_inode_cap(inode, cap, &invalidate);
2541 }
2542 spin_unlock(&ci->i_ceph_lock);
2543
2544 if (invalidate)
2545 ceph_queue_invalidate(inode);
2546 while (iputs--)
2547 iput(inode);
2548}