dma40: combine duplicated code in log_sg_to_dev
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / ceph / inode.c
blobe835eff551e342156ff1f1b1f1e657c089f435eb
1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/module.h>
4 #include <linux/fs.h>
5 #include <linux/slab.h>
6 #include <linux/string.h>
7 #include <linux/uaccess.h>
8 #include <linux/kernel.h>
9 #include <linux/namei.h>
10 #include <linux/writeback.h>
11 #include <linux/vmalloc.h>
12 #include <linux/pagevec.h>
14 #include "super.h"
15 #include "mds_client.h"
16 #include <linux/ceph/decode.h>
19 * Ceph inode operations
21 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
22 * setattr, etc.), xattr helpers, and helpers for assimilating
23 * metadata returned by the MDS into our cache.
25 * Also define helpers for doing asynchronous writeback, invalidation,
26 * and truncation for the benefit of those who can't afford to block
27 * (typically because they are in the message handler path).
30 static const struct inode_operations ceph_symlink_iops;
32 static void ceph_invalidate_work(struct work_struct *work);
33 static void ceph_writeback_work(struct work_struct *work);
34 static void ceph_vmtruncate_work(struct work_struct *work);
37 * find or create an inode, given the ceph ino number
39 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
41 struct inode *inode;
42 ino_t t = ceph_vino_to_ino(vino);
44 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
45 if (inode == NULL)
46 return ERR_PTR(-ENOMEM);
47 if (inode->i_state & I_NEW) {
48 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
49 inode, ceph_vinop(inode), (u64)inode->i_ino);
50 unlock_new_inode(inode);
53 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
54 vino.snap, inode);
55 return inode;
59 * get/constuct snapdir inode for a given directory
61 struct inode *ceph_get_snapdir(struct inode *parent)
63 struct ceph_vino vino = {
64 .ino = ceph_ino(parent),
65 .snap = CEPH_SNAPDIR,
67 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
68 struct ceph_inode_info *ci = ceph_inode(inode);
70 BUG_ON(!S_ISDIR(parent->i_mode));
71 if (IS_ERR(inode))
72 return inode;
73 inode->i_mode = parent->i_mode;
74 inode->i_uid = parent->i_uid;
75 inode->i_gid = parent->i_gid;
76 inode->i_op = &ceph_dir_iops;
77 inode->i_fop = &ceph_dir_fops;
78 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
79 ci->i_rbytes = 0;
80 return inode;
83 const struct inode_operations ceph_file_iops = {
84 .permission = ceph_permission,
85 .setattr = ceph_setattr,
86 .getattr = ceph_getattr,
87 .setxattr = ceph_setxattr,
88 .getxattr = ceph_getxattr,
89 .listxattr = ceph_listxattr,
90 .removexattr = ceph_removexattr,
95 * We use a 'frag tree' to keep track of the MDS's directory fragments
96 * for a given inode (usually there is just a single fragment). We
97 * need to know when a child frag is delegated to a new MDS, or when
98 * it is flagged as replicated, so we can direct our requests
99 * accordingly.
103 * find/create a frag in the tree
105 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
106 u32 f)
108 struct rb_node **p;
109 struct rb_node *parent = NULL;
110 struct ceph_inode_frag *frag;
111 int c;
113 p = &ci->i_fragtree.rb_node;
114 while (*p) {
115 parent = *p;
116 frag = rb_entry(parent, struct ceph_inode_frag, node);
117 c = ceph_frag_compare(f, frag->frag);
118 if (c < 0)
119 p = &(*p)->rb_left;
120 else if (c > 0)
121 p = &(*p)->rb_right;
122 else
123 return frag;
126 frag = kmalloc(sizeof(*frag), GFP_NOFS);
127 if (!frag) {
128 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
129 "frag %x\n", &ci->vfs_inode,
130 ceph_vinop(&ci->vfs_inode), f);
131 return ERR_PTR(-ENOMEM);
133 frag->frag = f;
134 frag->split_by = 0;
135 frag->mds = -1;
136 frag->ndist = 0;
138 rb_link_node(&frag->node, parent, p);
139 rb_insert_color(&frag->node, &ci->i_fragtree);
141 dout("get_or_create_frag added %llx.%llx frag %x\n",
142 ceph_vinop(&ci->vfs_inode), f);
143 return frag;
147 * find a specific frag @f
149 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
151 struct rb_node *n = ci->i_fragtree.rb_node;
153 while (n) {
154 struct ceph_inode_frag *frag =
155 rb_entry(n, struct ceph_inode_frag, node);
156 int c = ceph_frag_compare(f, frag->frag);
157 if (c < 0)
158 n = n->rb_left;
159 else if (c > 0)
160 n = n->rb_right;
161 else
162 return frag;
164 return NULL;
168 * Choose frag containing the given value @v. If @pfrag is
169 * specified, copy the frag delegation info to the caller if
170 * it is present.
172 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
173 struct ceph_inode_frag *pfrag,
174 int *found)
176 u32 t = ceph_frag_make(0, 0);
177 struct ceph_inode_frag *frag;
178 unsigned nway, i;
179 u32 n;
181 if (found)
182 *found = 0;
184 mutex_lock(&ci->i_fragtree_mutex);
185 while (1) {
186 WARN_ON(!ceph_frag_contains_value(t, v));
187 frag = __ceph_find_frag(ci, t);
188 if (!frag)
189 break; /* t is a leaf */
190 if (frag->split_by == 0) {
191 if (pfrag)
192 memcpy(pfrag, frag, sizeof(*pfrag));
193 if (found)
194 *found = 1;
195 break;
198 /* choose child */
199 nway = 1 << frag->split_by;
200 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
201 frag->split_by, nway);
202 for (i = 0; i < nway; i++) {
203 n = ceph_frag_make_child(t, frag->split_by, i);
204 if (ceph_frag_contains_value(n, v)) {
205 t = n;
206 break;
209 BUG_ON(i == nway);
211 dout("choose_frag(%x) = %x\n", v, t);
213 mutex_unlock(&ci->i_fragtree_mutex);
214 return t;
218 * Process dirfrag (delegation) info from the mds. Include leaf
219 * fragment in tree ONLY if ndist > 0. Otherwise, only
220 * branches/splits are included in i_fragtree)
222 static int ceph_fill_dirfrag(struct inode *inode,
223 struct ceph_mds_reply_dirfrag *dirinfo)
225 struct ceph_inode_info *ci = ceph_inode(inode);
226 struct ceph_inode_frag *frag;
227 u32 id = le32_to_cpu(dirinfo->frag);
228 int mds = le32_to_cpu(dirinfo->auth);
229 int ndist = le32_to_cpu(dirinfo->ndist);
230 int i;
231 int err = 0;
233 mutex_lock(&ci->i_fragtree_mutex);
234 if (ndist == 0) {
235 /* no delegation info needed. */
236 frag = __ceph_find_frag(ci, id);
237 if (!frag)
238 goto out;
239 if (frag->split_by == 0) {
240 /* tree leaf, remove */
241 dout("fill_dirfrag removed %llx.%llx frag %x"
242 " (no ref)\n", ceph_vinop(inode), id);
243 rb_erase(&frag->node, &ci->i_fragtree);
244 kfree(frag);
245 } else {
246 /* tree branch, keep and clear */
247 dout("fill_dirfrag cleared %llx.%llx frag %x"
248 " referral\n", ceph_vinop(inode), id);
249 frag->mds = -1;
250 frag->ndist = 0;
252 goto out;
256 /* find/add this frag to store mds delegation info */
257 frag = __get_or_create_frag(ci, id);
258 if (IS_ERR(frag)) {
259 /* this is not the end of the world; we can continue
260 with bad/inaccurate delegation info */
261 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
262 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
263 err = -ENOMEM;
264 goto out;
267 frag->mds = mds;
268 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
269 for (i = 0; i < frag->ndist; i++)
270 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
271 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
272 ceph_vinop(inode), frag->frag, frag->ndist);
274 out:
275 mutex_unlock(&ci->i_fragtree_mutex);
276 return err;
281 * initialize a newly allocated inode.
283 struct inode *ceph_alloc_inode(struct super_block *sb)
285 struct ceph_inode_info *ci;
286 int i;
288 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
289 if (!ci)
290 return NULL;
292 dout("alloc_inode %p\n", &ci->vfs_inode);
294 ci->i_version = 0;
295 ci->i_time_warp_seq = 0;
296 ci->i_ceph_flags = 0;
297 ci->i_release_count = 0;
298 ci->i_symlink = NULL;
300 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
302 ci->i_fragtree = RB_ROOT;
303 mutex_init(&ci->i_fragtree_mutex);
305 ci->i_xattrs.blob = NULL;
306 ci->i_xattrs.prealloc_blob = NULL;
307 ci->i_xattrs.dirty = false;
308 ci->i_xattrs.index = RB_ROOT;
309 ci->i_xattrs.count = 0;
310 ci->i_xattrs.names_size = 0;
311 ci->i_xattrs.vals_size = 0;
312 ci->i_xattrs.version = 0;
313 ci->i_xattrs.index_version = 0;
315 ci->i_caps = RB_ROOT;
316 ci->i_auth_cap = NULL;
317 ci->i_dirty_caps = 0;
318 ci->i_flushing_caps = 0;
319 INIT_LIST_HEAD(&ci->i_dirty_item);
320 INIT_LIST_HEAD(&ci->i_flushing_item);
321 ci->i_cap_flush_seq = 0;
322 ci->i_cap_flush_last_tid = 0;
323 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
324 init_waitqueue_head(&ci->i_cap_wq);
325 ci->i_hold_caps_min = 0;
326 ci->i_hold_caps_max = 0;
327 INIT_LIST_HEAD(&ci->i_cap_delay_list);
328 ci->i_cap_exporting_mds = 0;
329 ci->i_cap_exporting_mseq = 0;
330 ci->i_cap_exporting_issued = 0;
331 INIT_LIST_HEAD(&ci->i_cap_snaps);
332 ci->i_head_snapc = NULL;
333 ci->i_snap_caps = 0;
335 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
336 ci->i_nr_by_mode[i] = 0;
338 ci->i_truncate_seq = 0;
339 ci->i_truncate_size = 0;
340 ci->i_truncate_pending = 0;
342 ci->i_max_size = 0;
343 ci->i_reported_size = 0;
344 ci->i_wanted_max_size = 0;
345 ci->i_requested_max_size = 0;
347 ci->i_pin_ref = 0;
348 ci->i_rd_ref = 0;
349 ci->i_rdcache_ref = 0;
350 ci->i_wr_ref = 0;
351 ci->i_wrbuffer_ref = 0;
352 ci->i_wrbuffer_ref_head = 0;
353 ci->i_shared_gen = 0;
354 ci->i_rdcache_gen = 0;
355 ci->i_rdcache_revoking = 0;
357 INIT_LIST_HEAD(&ci->i_unsafe_writes);
358 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
359 spin_lock_init(&ci->i_unsafe_lock);
361 ci->i_snap_realm = NULL;
362 INIT_LIST_HEAD(&ci->i_snap_realm_item);
363 INIT_LIST_HEAD(&ci->i_snap_flush_item);
365 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
366 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
368 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
370 return &ci->vfs_inode;
373 static void ceph_i_callback(struct rcu_head *head)
375 struct inode *inode = container_of(head, struct inode, i_rcu);
376 struct ceph_inode_info *ci = ceph_inode(inode);
378 INIT_LIST_HEAD(&inode->i_dentry);
379 kmem_cache_free(ceph_inode_cachep, ci);
382 void ceph_destroy_inode(struct inode *inode)
384 struct ceph_inode_info *ci = ceph_inode(inode);
385 struct ceph_inode_frag *frag;
386 struct rb_node *n;
388 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
390 ceph_queue_caps_release(inode);
393 * we may still have a snap_realm reference if there are stray
394 * caps in i_cap_exporting_issued or i_snap_caps.
396 if (ci->i_snap_realm) {
397 struct ceph_mds_client *mdsc =
398 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
399 struct ceph_snap_realm *realm = ci->i_snap_realm;
401 dout(" dropping residual ref to snap realm %p\n", realm);
402 spin_lock(&realm->inodes_with_caps_lock);
403 list_del_init(&ci->i_snap_realm_item);
404 spin_unlock(&realm->inodes_with_caps_lock);
405 ceph_put_snap_realm(mdsc, realm);
408 kfree(ci->i_symlink);
409 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
410 frag = rb_entry(n, struct ceph_inode_frag, node);
411 rb_erase(n, &ci->i_fragtree);
412 kfree(frag);
415 __ceph_destroy_xattrs(ci);
416 if (ci->i_xattrs.blob)
417 ceph_buffer_put(ci->i_xattrs.blob);
418 if (ci->i_xattrs.prealloc_blob)
419 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
421 call_rcu(&inode->i_rcu, ceph_i_callback);
426 * Helpers to fill in size, ctime, mtime, and atime. We have to be
427 * careful because either the client or MDS may have more up to date
428 * info, depending on which capabilities are held, and whether
429 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
430 * and size are monotonically increasing, except when utimes() or
431 * truncate() increments the corresponding _seq values.)
433 int ceph_fill_file_size(struct inode *inode, int issued,
434 u32 truncate_seq, u64 truncate_size, u64 size)
436 struct ceph_inode_info *ci = ceph_inode(inode);
437 int queue_trunc = 0;
439 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
440 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
441 dout("size %lld -> %llu\n", inode->i_size, size);
442 inode->i_size = size;
443 inode->i_blocks = (size + (1<<9) - 1) >> 9;
444 ci->i_reported_size = size;
445 if (truncate_seq != ci->i_truncate_seq) {
446 dout("truncate_seq %u -> %u\n",
447 ci->i_truncate_seq, truncate_seq);
448 ci->i_truncate_seq = truncate_seq;
450 * If we hold relevant caps, or in the case where we're
451 * not the only client referencing this file and we
452 * don't hold those caps, then we need to check whether
453 * the file is either opened or mmaped
455 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
456 CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
457 CEPH_CAP_FILE_EXCL|
458 CEPH_CAP_FILE_LAZYIO)) ||
459 mapping_mapped(inode->i_mapping) ||
460 __ceph_caps_file_wanted(ci)) {
461 ci->i_truncate_pending++;
462 queue_trunc = 1;
466 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
467 ci->i_truncate_size != truncate_size) {
468 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
469 truncate_size);
470 ci->i_truncate_size = truncate_size;
472 return queue_trunc;
475 void ceph_fill_file_time(struct inode *inode, int issued,
476 u64 time_warp_seq, struct timespec *ctime,
477 struct timespec *mtime, struct timespec *atime)
479 struct ceph_inode_info *ci = ceph_inode(inode);
480 int warn = 0;
482 if (issued & (CEPH_CAP_FILE_EXCL|
483 CEPH_CAP_FILE_WR|
484 CEPH_CAP_FILE_BUFFER|
485 CEPH_CAP_AUTH_EXCL|
486 CEPH_CAP_XATTR_EXCL)) {
487 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
488 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
489 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
490 ctime->tv_sec, ctime->tv_nsec);
491 inode->i_ctime = *ctime;
493 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
494 /* the MDS did a utimes() */
495 dout("mtime %ld.%09ld -> %ld.%09ld "
496 "tw %d -> %d\n",
497 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
498 mtime->tv_sec, mtime->tv_nsec,
499 ci->i_time_warp_seq, (int)time_warp_seq);
501 inode->i_mtime = *mtime;
502 inode->i_atime = *atime;
503 ci->i_time_warp_seq = time_warp_seq;
504 } else if (time_warp_seq == ci->i_time_warp_seq) {
505 /* nobody did utimes(); take the max */
506 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
507 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
508 inode->i_mtime.tv_sec,
509 inode->i_mtime.tv_nsec,
510 mtime->tv_sec, mtime->tv_nsec);
511 inode->i_mtime = *mtime;
513 if (timespec_compare(atime, &inode->i_atime) > 0) {
514 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
515 inode->i_atime.tv_sec,
516 inode->i_atime.tv_nsec,
517 atime->tv_sec, atime->tv_nsec);
518 inode->i_atime = *atime;
520 } else if (issued & CEPH_CAP_FILE_EXCL) {
521 /* we did a utimes(); ignore mds values */
522 } else {
523 warn = 1;
525 } else {
526 /* we have no write|excl caps; whatever the MDS says is true */
527 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
528 inode->i_ctime = *ctime;
529 inode->i_mtime = *mtime;
530 inode->i_atime = *atime;
531 ci->i_time_warp_seq = time_warp_seq;
532 } else {
533 warn = 1;
536 if (warn) /* time_warp_seq shouldn't go backwards */
537 dout("%p mds time_warp_seq %llu < %u\n",
538 inode, time_warp_seq, ci->i_time_warp_seq);
542 * Populate an inode based on info from mds. May be called on new or
543 * existing inodes.
545 static int fill_inode(struct inode *inode,
546 struct ceph_mds_reply_info_in *iinfo,
547 struct ceph_mds_reply_dirfrag *dirinfo,
548 struct ceph_mds_session *session,
549 unsigned long ttl_from, int cap_fmode,
550 struct ceph_cap_reservation *caps_reservation)
552 struct ceph_mds_reply_inode *info = iinfo->in;
553 struct ceph_inode_info *ci = ceph_inode(inode);
554 int i;
555 int issued, implemented;
556 struct timespec mtime, atime, ctime;
557 u32 nsplits;
558 struct ceph_buffer *xattr_blob = NULL;
559 int err = 0;
560 int queue_trunc = 0;
562 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
563 inode, ceph_vinop(inode), le64_to_cpu(info->version),
564 ci->i_version);
567 * prealloc xattr data, if it looks like we'll need it. only
568 * if len > 4 (meaning there are actually xattrs; the first 4
569 * bytes are the xattr count).
571 if (iinfo->xattr_len > 4) {
572 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
573 if (!xattr_blob)
574 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
575 iinfo->xattr_len);
578 spin_lock(&inode->i_lock);
581 * provided version will be odd if inode value is projected,
582 * even if stable. skip the update if we have newer stable
583 * info (ours>=theirs, e.g. due to racing mds replies), unless
584 * we are getting projected (unstable) info (in which case the
585 * version is odd, and we want ours>theirs).
586 * us them
587 * 2 2 skip
588 * 3 2 skip
589 * 3 3 update
591 if (le64_to_cpu(info->version) > 0 &&
592 (ci->i_version & ~1) >= le64_to_cpu(info->version))
593 goto no_change;
595 issued = __ceph_caps_issued(ci, &implemented);
596 issued |= implemented | __ceph_caps_dirty(ci);
598 /* update inode */
599 ci->i_version = le64_to_cpu(info->version);
600 inode->i_version++;
601 inode->i_rdev = le32_to_cpu(info->rdev);
603 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
604 inode->i_mode = le32_to_cpu(info->mode);
605 inode->i_uid = le32_to_cpu(info->uid);
606 inode->i_gid = le32_to_cpu(info->gid);
607 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
608 inode->i_uid, inode->i_gid);
611 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
612 inode->i_nlink = le32_to_cpu(info->nlink);
614 /* be careful with mtime, atime, size */
615 ceph_decode_timespec(&atime, &info->atime);
616 ceph_decode_timespec(&mtime, &info->mtime);
617 ceph_decode_timespec(&ctime, &info->ctime);
618 queue_trunc = ceph_fill_file_size(inode, issued,
619 le32_to_cpu(info->truncate_seq),
620 le64_to_cpu(info->truncate_size),
621 le64_to_cpu(info->size));
622 ceph_fill_file_time(inode, issued,
623 le32_to_cpu(info->time_warp_seq),
624 &ctime, &mtime, &atime);
626 /* only update max_size on auth cap */
627 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
628 ci->i_max_size != le64_to_cpu(info->max_size)) {
629 dout("max_size %lld -> %llu\n", ci->i_max_size,
630 le64_to_cpu(info->max_size));
631 ci->i_max_size = le64_to_cpu(info->max_size);
634 ci->i_layout = info->layout;
635 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
637 /* xattrs */
638 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
639 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
640 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
641 if (ci->i_xattrs.blob)
642 ceph_buffer_put(ci->i_xattrs.blob);
643 ci->i_xattrs.blob = xattr_blob;
644 if (xattr_blob)
645 memcpy(ci->i_xattrs.blob->vec.iov_base,
646 iinfo->xattr_data, iinfo->xattr_len);
647 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
648 xattr_blob = NULL;
651 inode->i_mapping->a_ops = &ceph_aops;
652 inode->i_mapping->backing_dev_info =
653 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
655 switch (inode->i_mode & S_IFMT) {
656 case S_IFIFO:
657 case S_IFBLK:
658 case S_IFCHR:
659 case S_IFSOCK:
660 init_special_inode(inode, inode->i_mode, inode->i_rdev);
661 inode->i_op = &ceph_file_iops;
662 break;
663 case S_IFREG:
664 inode->i_op = &ceph_file_iops;
665 inode->i_fop = &ceph_file_fops;
666 break;
667 case S_IFLNK:
668 inode->i_op = &ceph_symlink_iops;
669 if (!ci->i_symlink) {
670 int symlen = iinfo->symlink_len;
671 char *sym;
673 BUG_ON(symlen != inode->i_size);
674 spin_unlock(&inode->i_lock);
676 err = -ENOMEM;
677 sym = kmalloc(symlen+1, GFP_NOFS);
678 if (!sym)
679 goto out;
680 memcpy(sym, iinfo->symlink, symlen);
681 sym[symlen] = 0;
683 spin_lock(&inode->i_lock);
684 if (!ci->i_symlink)
685 ci->i_symlink = sym;
686 else
687 kfree(sym); /* lost a race */
689 break;
690 case S_IFDIR:
691 inode->i_op = &ceph_dir_iops;
692 inode->i_fop = &ceph_dir_fops;
694 ci->i_dir_layout = iinfo->dir_layout;
696 ci->i_files = le64_to_cpu(info->files);
697 ci->i_subdirs = le64_to_cpu(info->subdirs);
698 ci->i_rbytes = le64_to_cpu(info->rbytes);
699 ci->i_rfiles = le64_to_cpu(info->rfiles);
700 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
701 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
703 /* set dir completion flag? */
704 if (ci->i_files == 0 && ci->i_subdirs == 0 &&
705 ceph_snap(inode) == CEPH_NOSNAP &&
706 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
707 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
708 (ci->i_ceph_flags & CEPH_I_COMPLETE) == 0) {
709 dout(" marking %p complete (empty)\n", inode);
710 ci->i_ceph_flags |= CEPH_I_COMPLETE;
711 ci->i_max_offset = 2;
714 /* it may be better to set st_size in getattr instead? */
715 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), RBYTES))
716 inode->i_size = ci->i_rbytes;
717 break;
718 default:
719 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
720 ceph_vinop(inode), inode->i_mode);
723 no_change:
724 spin_unlock(&inode->i_lock);
726 /* queue truncate if we saw i_size decrease */
727 if (queue_trunc)
728 ceph_queue_vmtruncate(inode);
730 /* populate frag tree */
731 /* FIXME: move me up, if/when version reflects fragtree changes */
732 nsplits = le32_to_cpu(info->fragtree.nsplits);
733 mutex_lock(&ci->i_fragtree_mutex);
734 for (i = 0; i < nsplits; i++) {
735 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
736 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
738 if (IS_ERR(frag))
739 continue;
740 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
741 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
743 mutex_unlock(&ci->i_fragtree_mutex);
745 /* were we issued a capability? */
746 if (info->cap.caps) {
747 if (ceph_snap(inode) == CEPH_NOSNAP) {
748 ceph_add_cap(inode, session,
749 le64_to_cpu(info->cap.cap_id),
750 cap_fmode,
751 le32_to_cpu(info->cap.caps),
752 le32_to_cpu(info->cap.wanted),
753 le32_to_cpu(info->cap.seq),
754 le32_to_cpu(info->cap.mseq),
755 le64_to_cpu(info->cap.realm),
756 info->cap.flags,
757 caps_reservation);
758 } else {
759 spin_lock(&inode->i_lock);
760 dout(" %p got snap_caps %s\n", inode,
761 ceph_cap_string(le32_to_cpu(info->cap.caps)));
762 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
763 if (cap_fmode >= 0)
764 __ceph_get_fmode(ci, cap_fmode);
765 spin_unlock(&inode->i_lock);
767 } else if (cap_fmode >= 0) {
768 pr_warning("mds issued no caps on %llx.%llx\n",
769 ceph_vinop(inode));
770 __ceph_get_fmode(ci, cap_fmode);
773 /* update delegation info? */
774 if (dirinfo)
775 ceph_fill_dirfrag(inode, dirinfo);
777 err = 0;
779 out:
780 if (xattr_blob)
781 ceph_buffer_put(xattr_blob);
782 return err;
786 * caller should hold session s_mutex.
788 static void update_dentry_lease(struct dentry *dentry,
789 struct ceph_mds_reply_lease *lease,
790 struct ceph_mds_session *session,
791 unsigned long from_time)
793 struct ceph_dentry_info *di = ceph_dentry(dentry);
794 long unsigned duration = le32_to_cpu(lease->duration_ms);
795 long unsigned ttl = from_time + (duration * HZ) / 1000;
796 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
797 struct inode *dir;
799 /* only track leases on regular dentries */
800 if (dentry->d_op != &ceph_dentry_ops)
801 return;
803 spin_lock(&dentry->d_lock);
804 dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n",
805 dentry, le16_to_cpu(lease->mask), duration, ttl);
807 /* make lease_rdcache_gen match directory */
808 dir = dentry->d_parent->d_inode;
809 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
811 if (lease->mask == 0)
812 goto out_unlock;
814 if (di->lease_gen == session->s_cap_gen &&
815 time_before(ttl, dentry->d_time))
816 goto out_unlock; /* we already have a newer lease. */
818 if (di->lease_session && di->lease_session != session)
819 goto out_unlock;
821 ceph_dentry_lru_touch(dentry);
823 if (!di->lease_session)
824 di->lease_session = ceph_get_mds_session(session);
825 di->lease_gen = session->s_cap_gen;
826 di->lease_seq = le32_to_cpu(lease->seq);
827 di->lease_renew_after = half_ttl;
828 di->lease_renew_from = 0;
829 dentry->d_time = ttl;
830 out_unlock:
831 spin_unlock(&dentry->d_lock);
832 return;
836 * Set dentry's directory position based on the current dir's max, and
837 * order it in d_subdirs, so that dcache_readdir behaves.
839 static void ceph_set_dentry_offset(struct dentry *dn)
841 struct dentry *dir = dn->d_parent;
842 struct inode *inode = dn->d_parent->d_inode;
843 struct ceph_dentry_info *di;
845 BUG_ON(!inode);
847 di = ceph_dentry(dn);
849 spin_lock(&inode->i_lock);
850 if ((ceph_inode(inode)->i_ceph_flags & CEPH_I_COMPLETE) == 0) {
851 spin_unlock(&inode->i_lock);
852 return;
854 di->offset = ceph_inode(inode)->i_max_offset++;
855 spin_unlock(&inode->i_lock);
857 spin_lock(&dir->d_lock);
858 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
859 list_move(&dn->d_u.d_child, &dir->d_subdirs);
860 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
861 dn->d_u.d_child.prev, dn->d_u.d_child.next);
862 spin_unlock(&dn->d_lock);
863 spin_unlock(&dir->d_lock);
867 * splice a dentry to an inode.
868 * caller must hold directory i_mutex for this to be safe.
870 * we will only rehash the resulting dentry if @prehash is
871 * true; @prehash will be set to false (for the benefit of
872 * the caller) if we fail.
874 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
875 bool *prehash, bool set_offset)
877 struct dentry *realdn;
879 BUG_ON(dn->d_inode);
881 /* dn must be unhashed */
882 if (!d_unhashed(dn))
883 d_drop(dn);
884 realdn = d_materialise_unique(dn, in);
885 if (IS_ERR(realdn)) {
886 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
887 PTR_ERR(realdn), dn, in, ceph_vinop(in));
888 if (prehash)
889 *prehash = false; /* don't rehash on error */
890 dn = realdn; /* note realdn contains the error */
891 goto out;
892 } else if (realdn) {
893 dout("dn %p (%d) spliced with %p (%d) "
894 "inode %p ino %llx.%llx\n",
895 dn, dn->d_count,
896 realdn, realdn->d_count,
897 realdn->d_inode, ceph_vinop(realdn->d_inode));
898 dput(dn);
899 dn = realdn;
900 } else {
901 BUG_ON(!ceph_dentry(dn));
902 dout("dn %p attached to %p ino %llx.%llx\n",
903 dn, dn->d_inode, ceph_vinop(dn->d_inode));
905 if ((!prehash || *prehash) && d_unhashed(dn))
906 d_rehash(dn);
907 if (set_offset)
908 ceph_set_dentry_offset(dn);
909 out:
910 return dn;
914 * Incorporate results into the local cache. This is either just
915 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
916 * after a lookup).
918 * A reply may contain
919 * a directory inode along with a dentry.
920 * and/or a target inode
922 * Called with snap_rwsem (read).
924 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
925 struct ceph_mds_session *session)
927 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
928 struct inode *in = NULL;
929 struct ceph_mds_reply_inode *ininfo;
930 struct ceph_vino vino;
931 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
932 int i = 0;
933 int err = 0;
935 dout("fill_trace %p is_dentry %d is_target %d\n", req,
936 rinfo->head->is_dentry, rinfo->head->is_target);
938 #if 0
940 * Debugging hook:
942 * If we resend completed ops to a recovering mds, we get no
943 * trace. Since that is very rare, pretend this is the case
944 * to ensure the 'no trace' handlers in the callers behave.
946 * Fill in inodes unconditionally to avoid breaking cap
947 * invariants.
949 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
950 pr_info("fill_trace faking empty trace on %lld %s\n",
951 req->r_tid, ceph_mds_op_name(rinfo->head->op));
952 if (rinfo->head->is_dentry) {
953 rinfo->head->is_dentry = 0;
954 err = fill_inode(req->r_locked_dir,
955 &rinfo->diri, rinfo->dirfrag,
956 session, req->r_request_started, -1);
958 if (rinfo->head->is_target) {
959 rinfo->head->is_target = 0;
960 ininfo = rinfo->targeti.in;
961 vino.ino = le64_to_cpu(ininfo->ino);
962 vino.snap = le64_to_cpu(ininfo->snapid);
963 in = ceph_get_inode(sb, vino);
964 err = fill_inode(in, &rinfo->targeti, NULL,
965 session, req->r_request_started,
966 req->r_fmode);
967 iput(in);
970 #endif
972 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
973 dout("fill_trace reply is empty!\n");
974 if (rinfo->head->result == 0 && req->r_locked_dir)
975 ceph_invalidate_dir_request(req);
976 return 0;
979 if (rinfo->head->is_dentry) {
980 struct inode *dir = req->r_locked_dir;
982 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
983 session, req->r_request_started, -1,
984 &req->r_caps_reservation);
985 if (err < 0)
986 return err;
990 * ignore null lease/binding on snapdir ENOENT, or else we
991 * will have trouble splicing in the virtual snapdir later
993 if (rinfo->head->is_dentry && !req->r_aborted &&
994 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
995 fsc->mount_options->snapdir_name,
996 req->r_dentry->d_name.len))) {
998 * lookup link rename : null -> possibly existing inode
999 * mknod symlink mkdir : null -> new inode
1000 * unlink : linked -> null
1002 struct inode *dir = req->r_locked_dir;
1003 struct dentry *dn = req->r_dentry;
1004 bool have_dir_cap, have_lease;
1006 BUG_ON(!dn);
1007 BUG_ON(!dir);
1008 BUG_ON(dn->d_parent->d_inode != dir);
1009 BUG_ON(ceph_ino(dir) !=
1010 le64_to_cpu(rinfo->diri.in->ino));
1011 BUG_ON(ceph_snap(dir) !=
1012 le64_to_cpu(rinfo->diri.in->snapid));
1014 /* do we have a lease on the whole dir? */
1015 have_dir_cap =
1016 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1017 CEPH_CAP_FILE_SHARED);
1019 /* do we have a dn lease? */
1020 have_lease = have_dir_cap ||
1021 (le16_to_cpu(rinfo->dlease->mask) &
1022 CEPH_LOCK_DN);
1024 if (!have_lease)
1025 dout("fill_trace no dentry lease or dir cap\n");
1027 /* rename? */
1028 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1029 dout(" src %p '%.*s' dst %p '%.*s'\n",
1030 req->r_old_dentry,
1031 req->r_old_dentry->d_name.len,
1032 req->r_old_dentry->d_name.name,
1033 dn, dn->d_name.len, dn->d_name.name);
1034 dout("fill_trace doing d_move %p -> %p\n",
1035 req->r_old_dentry, dn);
1037 /* d_move screws up d_subdirs order */
1038 ceph_i_clear(dir, CEPH_I_COMPLETE);
1040 d_move(req->r_old_dentry, dn);
1041 dout(" src %p '%.*s' dst %p '%.*s'\n",
1042 req->r_old_dentry,
1043 req->r_old_dentry->d_name.len,
1044 req->r_old_dentry->d_name.name,
1045 dn, dn->d_name.len, dn->d_name.name);
1047 /* ensure target dentry is invalidated, despite
1048 rehashing bug in vfs_rename_dir */
1049 ceph_invalidate_dentry_lease(dn);
1051 /* take overwritten dentry's readdir offset */
1052 dout("dn %p gets %p offset %lld (old offset %lld)\n",
1053 req->r_old_dentry, dn, ceph_dentry(dn)->offset,
1054 ceph_dentry(req->r_old_dentry)->offset);
1055 ceph_dentry(req->r_old_dentry)->offset =
1056 ceph_dentry(dn)->offset;
1058 dn = req->r_old_dentry; /* use old_dentry */
1059 in = dn->d_inode;
1062 /* null dentry? */
1063 if (!rinfo->head->is_target) {
1064 dout("fill_trace null dentry\n");
1065 if (dn->d_inode) {
1066 dout("d_delete %p\n", dn);
1067 d_delete(dn);
1068 } else {
1069 dout("d_instantiate %p NULL\n", dn);
1070 d_instantiate(dn, NULL);
1071 if (have_lease && d_unhashed(dn))
1072 d_rehash(dn);
1073 update_dentry_lease(dn, rinfo->dlease,
1074 session,
1075 req->r_request_started);
1077 goto done;
1080 /* attach proper inode */
1081 ininfo = rinfo->targeti.in;
1082 vino.ino = le64_to_cpu(ininfo->ino);
1083 vino.snap = le64_to_cpu(ininfo->snapid);
1084 in = dn->d_inode;
1085 if (!in) {
1086 in = ceph_get_inode(sb, vino);
1087 if (IS_ERR(in)) {
1088 pr_err("fill_trace bad get_inode "
1089 "%llx.%llx\n", vino.ino, vino.snap);
1090 err = PTR_ERR(in);
1091 d_delete(dn);
1092 goto done;
1094 dn = splice_dentry(dn, in, &have_lease, true);
1095 if (IS_ERR(dn)) {
1096 err = PTR_ERR(dn);
1097 goto done;
1099 req->r_dentry = dn; /* may have spliced */
1100 igrab(in);
1101 } else if (ceph_ino(in) == vino.ino &&
1102 ceph_snap(in) == vino.snap) {
1103 igrab(in);
1104 } else {
1105 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1106 dn, in, ceph_ino(in), ceph_snap(in),
1107 vino.ino, vino.snap);
1108 have_lease = false;
1109 in = NULL;
1112 if (have_lease)
1113 update_dentry_lease(dn, rinfo->dlease, session,
1114 req->r_request_started);
1115 dout(" final dn %p\n", dn);
1116 i++;
1117 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1118 req->r_op == CEPH_MDS_OP_MKSNAP) {
1119 struct dentry *dn = req->r_dentry;
1121 /* fill out a snapdir LOOKUPSNAP dentry */
1122 BUG_ON(!dn);
1123 BUG_ON(!req->r_locked_dir);
1124 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1125 ininfo = rinfo->targeti.in;
1126 vino.ino = le64_to_cpu(ininfo->ino);
1127 vino.snap = le64_to_cpu(ininfo->snapid);
1128 in = ceph_get_inode(sb, vino);
1129 if (IS_ERR(in)) {
1130 pr_err("fill_inode get_inode badness %llx.%llx\n",
1131 vino.ino, vino.snap);
1132 err = PTR_ERR(in);
1133 d_delete(dn);
1134 goto done;
1136 dout(" linking snapped dir %p to dn %p\n", in, dn);
1137 dn = splice_dentry(dn, in, NULL, true);
1138 if (IS_ERR(dn)) {
1139 err = PTR_ERR(dn);
1140 goto done;
1142 req->r_dentry = dn; /* may have spliced */
1143 igrab(in);
1144 rinfo->head->is_dentry = 1; /* fool notrace handlers */
1147 if (rinfo->head->is_target) {
1148 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1149 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1151 if (in == NULL || ceph_ino(in) != vino.ino ||
1152 ceph_snap(in) != vino.snap) {
1153 in = ceph_get_inode(sb, vino);
1154 if (IS_ERR(in)) {
1155 err = PTR_ERR(in);
1156 goto done;
1159 req->r_target_inode = in;
1161 err = fill_inode(in,
1162 &rinfo->targeti, NULL,
1163 session, req->r_request_started,
1164 (le32_to_cpu(rinfo->head->result) == 0) ?
1165 req->r_fmode : -1,
1166 &req->r_caps_reservation);
1167 if (err < 0) {
1168 pr_err("fill_inode badness %p %llx.%llx\n",
1169 in, ceph_vinop(in));
1170 goto done;
1174 done:
1175 dout("fill_trace done err=%d\n", err);
1176 return err;
1180 * Prepopulate our cache with readdir results, leases, etc.
1182 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1183 struct ceph_mds_session *session)
1185 struct dentry *parent = req->r_dentry;
1186 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1187 struct qstr dname;
1188 struct dentry *dn;
1189 struct inode *in;
1190 int err = 0, i;
1191 struct inode *snapdir = NULL;
1192 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1193 u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1194 struct ceph_dentry_info *di;
1196 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1197 snapdir = ceph_get_snapdir(parent->d_inode);
1198 parent = d_find_alias(snapdir);
1199 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1200 rinfo->dir_nr, parent);
1201 } else {
1202 dout("readdir_prepopulate %d items under dn %p\n",
1203 rinfo->dir_nr, parent);
1204 if (rinfo->dir_dir)
1205 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1208 for (i = 0; i < rinfo->dir_nr; i++) {
1209 struct ceph_vino vino;
1211 dname.name = rinfo->dir_dname[i];
1212 dname.len = rinfo->dir_dname_len[i];
1213 dname.hash = full_name_hash(dname.name, dname.len);
1215 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1216 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1218 retry_lookup:
1219 dn = d_lookup(parent, &dname);
1220 dout("d_lookup on parent=%p name=%.*s got %p\n",
1221 parent, dname.len, dname.name, dn);
1223 if (!dn) {
1224 dn = d_alloc(parent, &dname);
1225 dout("d_alloc %p '%.*s' = %p\n", parent,
1226 dname.len, dname.name, dn);
1227 if (dn == NULL) {
1228 dout("d_alloc badness\n");
1229 err = -ENOMEM;
1230 goto out;
1232 err = ceph_init_dentry(dn);
1233 if (err < 0) {
1234 dput(dn);
1235 goto out;
1237 } else if (dn->d_inode &&
1238 (ceph_ino(dn->d_inode) != vino.ino ||
1239 ceph_snap(dn->d_inode) != vino.snap)) {
1240 dout(" dn %p points to wrong inode %p\n",
1241 dn, dn->d_inode);
1242 d_delete(dn);
1243 dput(dn);
1244 goto retry_lookup;
1245 } else {
1246 /* reorder parent's d_subdirs */
1247 spin_lock(&parent->d_lock);
1248 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1249 list_move(&dn->d_u.d_child, &parent->d_subdirs);
1250 spin_unlock(&dn->d_lock);
1251 spin_unlock(&parent->d_lock);
1254 di = dn->d_fsdata;
1255 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1257 /* inode */
1258 if (dn->d_inode) {
1259 in = dn->d_inode;
1260 } else {
1261 in = ceph_get_inode(parent->d_sb, vino);
1262 if (IS_ERR(in)) {
1263 dout("new_inode badness\n");
1264 d_delete(dn);
1265 dput(dn);
1266 err = PTR_ERR(in);
1267 goto out;
1269 dn = splice_dentry(dn, in, NULL, false);
1270 if (IS_ERR(dn))
1271 dn = NULL;
1274 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1275 req->r_request_started, -1,
1276 &req->r_caps_reservation) < 0) {
1277 pr_err("fill_inode badness on %p\n", in);
1278 goto next_item;
1280 if (dn)
1281 update_dentry_lease(dn, rinfo->dir_dlease[i],
1282 req->r_session,
1283 req->r_request_started);
1284 next_item:
1285 if (dn)
1286 dput(dn);
1288 req->r_did_prepopulate = true;
1290 out:
1291 if (snapdir) {
1292 iput(snapdir);
1293 dput(parent);
1295 dout("readdir_prepopulate done\n");
1296 return err;
1299 int ceph_inode_set_size(struct inode *inode, loff_t size)
1301 struct ceph_inode_info *ci = ceph_inode(inode);
1302 int ret = 0;
1304 spin_lock(&inode->i_lock);
1305 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1306 inode->i_size = size;
1307 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1309 /* tell the MDS if we are approaching max_size */
1310 if ((size << 1) >= ci->i_max_size &&
1311 (ci->i_reported_size << 1) < ci->i_max_size)
1312 ret = 1;
1314 spin_unlock(&inode->i_lock);
1315 return ret;
1319 * Write back inode data in a worker thread. (This can't be done
1320 * in the message handler context.)
1322 void ceph_queue_writeback(struct inode *inode)
1324 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1325 &ceph_inode(inode)->i_wb_work)) {
1326 dout("ceph_queue_writeback %p\n", inode);
1327 igrab(inode);
1328 } else {
1329 dout("ceph_queue_writeback %p failed\n", inode);
1333 static void ceph_writeback_work(struct work_struct *work)
1335 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1336 i_wb_work);
1337 struct inode *inode = &ci->vfs_inode;
1339 dout("writeback %p\n", inode);
1340 filemap_fdatawrite(&inode->i_data);
1341 iput(inode);
1345 * queue an async invalidation
1347 void ceph_queue_invalidate(struct inode *inode)
1349 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1350 &ceph_inode(inode)->i_pg_inv_work)) {
1351 dout("ceph_queue_invalidate %p\n", inode);
1352 igrab(inode);
1353 } else {
1354 dout("ceph_queue_invalidate %p failed\n", inode);
1359 * invalidate any pages that are not dirty or under writeback. this
1360 * includes pages that are clean and mapped.
1362 static void ceph_invalidate_nondirty_pages(struct address_space *mapping)
1364 struct pagevec pvec;
1365 pgoff_t next = 0;
1366 int i;
1368 pagevec_init(&pvec, 0);
1369 while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
1370 for (i = 0; i < pagevec_count(&pvec); i++) {
1371 struct page *page = pvec.pages[i];
1372 pgoff_t index;
1373 int skip_page =
1374 (PageDirty(page) || PageWriteback(page));
1376 if (!skip_page)
1377 skip_page = !trylock_page(page);
1380 * We really shouldn't be looking at the ->index of an
1381 * unlocked page. But we're not allowed to lock these
1382 * pages. So we rely upon nobody altering the ->index
1383 * of this (pinned-by-us) page.
1385 index = page->index;
1386 if (index > next)
1387 next = index;
1388 next++;
1390 if (skip_page)
1391 continue;
1393 generic_error_remove_page(mapping, page);
1394 unlock_page(page);
1396 pagevec_release(&pvec);
1397 cond_resched();
1402 * Invalidate inode pages in a worker thread. (This can't be done
1403 * in the message handler context.)
1405 static void ceph_invalidate_work(struct work_struct *work)
1407 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1408 i_pg_inv_work);
1409 struct inode *inode = &ci->vfs_inode;
1410 u32 orig_gen;
1411 int check = 0;
1413 spin_lock(&inode->i_lock);
1414 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1415 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1416 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1417 /* nevermind! */
1418 spin_unlock(&inode->i_lock);
1419 goto out;
1421 orig_gen = ci->i_rdcache_gen;
1422 spin_unlock(&inode->i_lock);
1424 ceph_invalidate_nondirty_pages(inode->i_mapping);
1426 spin_lock(&inode->i_lock);
1427 if (orig_gen == ci->i_rdcache_gen &&
1428 orig_gen == ci->i_rdcache_revoking) {
1429 dout("invalidate_pages %p gen %d successful\n", inode,
1430 ci->i_rdcache_gen);
1431 ci->i_rdcache_revoking--;
1432 check = 1;
1433 } else {
1434 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1435 inode, orig_gen, ci->i_rdcache_gen,
1436 ci->i_rdcache_revoking);
1438 spin_unlock(&inode->i_lock);
1440 if (check)
1441 ceph_check_caps(ci, 0, NULL);
1442 out:
1443 iput(inode);
1448 * called by trunc_wq; take i_mutex ourselves
1450 * We also truncate in a separate thread as well.
1452 static void ceph_vmtruncate_work(struct work_struct *work)
1454 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1455 i_vmtruncate_work);
1456 struct inode *inode = &ci->vfs_inode;
1458 dout("vmtruncate_work %p\n", inode);
1459 mutex_lock(&inode->i_mutex);
1460 __ceph_do_pending_vmtruncate(inode);
1461 mutex_unlock(&inode->i_mutex);
1462 iput(inode);
1466 * Queue an async vmtruncate. If we fail to queue work, we will handle
1467 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1469 void ceph_queue_vmtruncate(struct inode *inode)
1471 struct ceph_inode_info *ci = ceph_inode(inode);
1473 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1474 &ci->i_vmtruncate_work)) {
1475 dout("ceph_queue_vmtruncate %p\n", inode);
1476 igrab(inode);
1477 } else {
1478 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1479 inode, ci->i_truncate_pending);
1484 * called with i_mutex held.
1486 * Make sure any pending truncation is applied before doing anything
1487 * that may depend on it.
1489 void __ceph_do_pending_vmtruncate(struct inode *inode)
1491 struct ceph_inode_info *ci = ceph_inode(inode);
1492 u64 to;
1493 int wrbuffer_refs, wake = 0;
1495 retry:
1496 spin_lock(&inode->i_lock);
1497 if (ci->i_truncate_pending == 0) {
1498 dout("__do_pending_vmtruncate %p none pending\n", inode);
1499 spin_unlock(&inode->i_lock);
1500 return;
1504 * make sure any dirty snapped pages are flushed before we
1505 * possibly truncate them.. so write AND block!
1507 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1508 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1509 inode);
1510 spin_unlock(&inode->i_lock);
1511 filemap_write_and_wait_range(&inode->i_data, 0,
1512 inode->i_sb->s_maxbytes);
1513 goto retry;
1516 to = ci->i_truncate_size;
1517 wrbuffer_refs = ci->i_wrbuffer_ref;
1518 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1519 ci->i_truncate_pending, to);
1520 spin_unlock(&inode->i_lock);
1522 truncate_inode_pages(inode->i_mapping, to);
1524 spin_lock(&inode->i_lock);
1525 ci->i_truncate_pending--;
1526 if (ci->i_truncate_pending == 0)
1527 wake = 1;
1528 spin_unlock(&inode->i_lock);
1530 if (wrbuffer_refs == 0)
1531 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1532 if (wake)
1533 wake_up_all(&ci->i_cap_wq);
1538 * symlinks
1540 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1542 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1543 nd_set_link(nd, ci->i_symlink);
1544 return NULL;
1547 static const struct inode_operations ceph_symlink_iops = {
1548 .readlink = generic_readlink,
1549 .follow_link = ceph_sym_follow_link,
1553 * setattr
1555 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1557 struct inode *inode = dentry->d_inode;
1558 struct ceph_inode_info *ci = ceph_inode(inode);
1559 struct inode *parent_inode = dentry->d_parent->d_inode;
1560 const unsigned int ia_valid = attr->ia_valid;
1561 struct ceph_mds_request *req;
1562 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1563 int issued;
1564 int release = 0, dirtied = 0;
1565 int mask = 0;
1566 int err = 0;
1568 if (ceph_snap(inode) != CEPH_NOSNAP)
1569 return -EROFS;
1571 __ceph_do_pending_vmtruncate(inode);
1573 err = inode_change_ok(inode, attr);
1574 if (err != 0)
1575 return err;
1577 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1578 USE_AUTH_MDS);
1579 if (IS_ERR(req))
1580 return PTR_ERR(req);
1582 spin_lock(&inode->i_lock);
1583 issued = __ceph_caps_issued(ci, NULL);
1584 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1586 if (ia_valid & ATTR_UID) {
1587 dout("setattr %p uid %d -> %d\n", inode,
1588 inode->i_uid, attr->ia_uid);
1589 if (issued & CEPH_CAP_AUTH_EXCL) {
1590 inode->i_uid = attr->ia_uid;
1591 dirtied |= CEPH_CAP_AUTH_EXCL;
1592 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1593 attr->ia_uid != inode->i_uid) {
1594 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid);
1595 mask |= CEPH_SETATTR_UID;
1596 release |= CEPH_CAP_AUTH_SHARED;
1599 if (ia_valid & ATTR_GID) {
1600 dout("setattr %p gid %d -> %d\n", inode,
1601 inode->i_gid, attr->ia_gid);
1602 if (issued & CEPH_CAP_AUTH_EXCL) {
1603 inode->i_gid = attr->ia_gid;
1604 dirtied |= CEPH_CAP_AUTH_EXCL;
1605 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1606 attr->ia_gid != inode->i_gid) {
1607 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid);
1608 mask |= CEPH_SETATTR_GID;
1609 release |= CEPH_CAP_AUTH_SHARED;
1612 if (ia_valid & ATTR_MODE) {
1613 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1614 attr->ia_mode);
1615 if (issued & CEPH_CAP_AUTH_EXCL) {
1616 inode->i_mode = attr->ia_mode;
1617 dirtied |= CEPH_CAP_AUTH_EXCL;
1618 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1619 attr->ia_mode != inode->i_mode) {
1620 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1621 mask |= CEPH_SETATTR_MODE;
1622 release |= CEPH_CAP_AUTH_SHARED;
1626 if (ia_valid & ATTR_ATIME) {
1627 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1628 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1629 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1630 if (issued & CEPH_CAP_FILE_EXCL) {
1631 ci->i_time_warp_seq++;
1632 inode->i_atime = attr->ia_atime;
1633 dirtied |= CEPH_CAP_FILE_EXCL;
1634 } else if ((issued & CEPH_CAP_FILE_WR) &&
1635 timespec_compare(&inode->i_atime,
1636 &attr->ia_atime) < 0) {
1637 inode->i_atime = attr->ia_atime;
1638 dirtied |= CEPH_CAP_FILE_WR;
1639 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1640 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1641 ceph_encode_timespec(&req->r_args.setattr.atime,
1642 &attr->ia_atime);
1643 mask |= CEPH_SETATTR_ATIME;
1644 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1645 CEPH_CAP_FILE_WR;
1648 if (ia_valid & ATTR_MTIME) {
1649 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1650 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1651 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1652 if (issued & CEPH_CAP_FILE_EXCL) {
1653 ci->i_time_warp_seq++;
1654 inode->i_mtime = attr->ia_mtime;
1655 dirtied |= CEPH_CAP_FILE_EXCL;
1656 } else if ((issued & CEPH_CAP_FILE_WR) &&
1657 timespec_compare(&inode->i_mtime,
1658 &attr->ia_mtime) < 0) {
1659 inode->i_mtime = attr->ia_mtime;
1660 dirtied |= CEPH_CAP_FILE_WR;
1661 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1662 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1663 ceph_encode_timespec(&req->r_args.setattr.mtime,
1664 &attr->ia_mtime);
1665 mask |= CEPH_SETATTR_MTIME;
1666 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1667 CEPH_CAP_FILE_WR;
1670 if (ia_valid & ATTR_SIZE) {
1671 dout("setattr %p size %lld -> %lld\n", inode,
1672 inode->i_size, attr->ia_size);
1673 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1674 err = -EINVAL;
1675 goto out;
1677 if ((issued & CEPH_CAP_FILE_EXCL) &&
1678 attr->ia_size > inode->i_size) {
1679 inode->i_size = attr->ia_size;
1680 inode->i_blocks =
1681 (attr->ia_size + (1 << 9) - 1) >> 9;
1682 inode->i_ctime = attr->ia_ctime;
1683 ci->i_reported_size = attr->ia_size;
1684 dirtied |= CEPH_CAP_FILE_EXCL;
1685 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1686 attr->ia_size != inode->i_size) {
1687 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1688 req->r_args.setattr.old_size =
1689 cpu_to_le64(inode->i_size);
1690 mask |= CEPH_SETATTR_SIZE;
1691 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1692 CEPH_CAP_FILE_WR;
1696 /* these do nothing */
1697 if (ia_valid & ATTR_CTIME) {
1698 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1699 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1700 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1701 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1702 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1703 only ? "ctime only" : "ignored");
1704 inode->i_ctime = attr->ia_ctime;
1705 if (only) {
1707 * if kernel wants to dirty ctime but nothing else,
1708 * we need to choose a cap to dirty under, or do
1709 * a almost-no-op setattr
1711 if (issued & CEPH_CAP_AUTH_EXCL)
1712 dirtied |= CEPH_CAP_AUTH_EXCL;
1713 else if (issued & CEPH_CAP_FILE_EXCL)
1714 dirtied |= CEPH_CAP_FILE_EXCL;
1715 else if (issued & CEPH_CAP_XATTR_EXCL)
1716 dirtied |= CEPH_CAP_XATTR_EXCL;
1717 else
1718 mask |= CEPH_SETATTR_CTIME;
1721 if (ia_valid & ATTR_FILE)
1722 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1724 if (dirtied) {
1725 __ceph_mark_dirty_caps(ci, dirtied);
1726 inode->i_ctime = CURRENT_TIME;
1729 release &= issued;
1730 spin_unlock(&inode->i_lock);
1732 if (mask) {
1733 req->r_inode = igrab(inode);
1734 req->r_inode_drop = release;
1735 req->r_args.setattr.mask = cpu_to_le32(mask);
1736 req->r_num_caps = 1;
1737 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1739 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1740 ceph_cap_string(dirtied), mask);
1742 ceph_mdsc_put_request(req);
1743 __ceph_do_pending_vmtruncate(inode);
1744 return err;
1745 out:
1746 spin_unlock(&inode->i_lock);
1747 ceph_mdsc_put_request(req);
1748 return err;
1752 * Verify that we have a lease on the given mask. If not,
1753 * do a getattr against an mds.
1755 int ceph_do_getattr(struct inode *inode, int mask)
1757 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1758 struct ceph_mds_client *mdsc = fsc->mdsc;
1759 struct ceph_mds_request *req;
1760 int err;
1762 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1763 dout("do_getattr inode %p SNAPDIR\n", inode);
1764 return 0;
1767 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1768 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1769 return 0;
1771 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1772 if (IS_ERR(req))
1773 return PTR_ERR(req);
1774 req->r_inode = igrab(inode);
1775 req->r_num_caps = 1;
1776 req->r_args.getattr.mask = cpu_to_le32(mask);
1777 err = ceph_mdsc_do_request(mdsc, NULL, req);
1778 ceph_mdsc_put_request(req);
1779 dout("do_getattr result=%d\n", err);
1780 return err;
1785 * Check inode permissions. We verify we have a valid value for
1786 * the AUTH cap, then call the generic handler.
1788 int ceph_permission(struct inode *inode, int mask, unsigned int flags)
1790 int err;
1792 if (flags & IPERM_FLAG_RCU)
1793 return -ECHILD;
1795 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1797 if (!err)
1798 err = generic_permission(inode, mask, flags, NULL);
1799 return err;
1803 * Get all attributes. Hopefully somedata we'll have a statlite()
1804 * and can limit the fields we require to be accurate.
1806 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1807 struct kstat *stat)
1809 struct inode *inode = dentry->d_inode;
1810 struct ceph_inode_info *ci = ceph_inode(inode);
1811 int err;
1813 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1814 if (!err) {
1815 generic_fillattr(inode, stat);
1816 stat->ino = inode->i_ino;
1817 if (ceph_snap(inode) != CEPH_NOSNAP)
1818 stat->dev = ceph_snap(inode);
1819 else
1820 stat->dev = 0;
1821 if (S_ISDIR(inode->i_mode)) {
1822 stat->size = ci->i_rbytes;
1823 stat->blocks = 0;
1824 stat->blksize = 65536;
1827 return err;