1
2#include <linux/ceph/ceph_debug.h>
3#include <linux/ceph/striper.h>
4
5#include <linux/module.h>
6#include <linux/sched.h>
7#include <linux/slab.h>
8#include <linux/file.h>
9#include <linux/mount.h>
10#include <linux/namei.h>
11#include <linux/writeback.h>
12#include <linux/falloc.h>
13#include <linux/iversion.h>
14#include <linux/ktime.h>
15
16#include "super.h"
17#include "mds_client.h"
18#include "cache.h"
19#include "io.h"
20#include "metric.h"
21
22static __le32 ceph_flags_sys2wire(u32 flags)
23{
24 u32 wire_flags = 0;
25
26 switch (flags & O_ACCMODE) {
27 case O_RDONLY:
28 wire_flags |= CEPH_O_RDONLY;
29 break;
30 case O_WRONLY:
31 wire_flags |= CEPH_O_WRONLY;
32 break;
33 case O_RDWR:
34 wire_flags |= CEPH_O_RDWR;
35 break;
36 }
37
38 flags &= ~O_ACCMODE;
39
40#define ceph_sys2wire(a) if (flags & a) { wire_flags |= CEPH_##a; flags &= ~a; }
41
42 ceph_sys2wire(O_CREAT);
43 ceph_sys2wire(O_EXCL);
44 ceph_sys2wire(O_TRUNC);
45 ceph_sys2wire(O_DIRECTORY);
46 ceph_sys2wire(O_NOFOLLOW);
47
48#undef ceph_sys2wire
49
50 if (flags)
51 dout("unused open flags: %x\n", flags);
52
53 return cpu_to_le32(wire_flags);
54}
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81#define ITER_GET_BVECS_PAGES 64
82
83static ssize_t __iter_get_bvecs(struct iov_iter *iter, size_t maxsize,
84 struct bio_vec *bvecs)
85{
86 size_t size = 0;
87 int bvec_idx = 0;
88
89 if (maxsize > iov_iter_count(iter))
90 maxsize = iov_iter_count(iter);
91
92 while (size < maxsize) {
93 struct page *pages[ITER_GET_BVECS_PAGES];
94 ssize_t bytes;
95 size_t start;
96 int idx = 0;
97
98 bytes = iov_iter_get_pages(iter, pages, maxsize - size,
99 ITER_GET_BVECS_PAGES, &start);
100 if (bytes < 0)
101 return size ?: bytes;
102
103 iov_iter_advance(iter, bytes);
104 size += bytes;
105
106 for ( ; bytes; idx++, bvec_idx++) {
107 struct bio_vec bv = {
108 .bv_page = pages[idx],
109 .bv_len = min_t(int, bytes, PAGE_SIZE - start),
110 .bv_offset = start,
111 };
112
113 bvecs[bvec_idx] = bv;
114 bytes -= bv.bv_len;
115 start = 0;
116 }
117 }
118
119 return size;
120}
121
122
123
124
125
126
127
128
129
130static ssize_t iter_get_bvecs_alloc(struct iov_iter *iter, size_t maxsize,
131 struct bio_vec **bvecs, int *num_bvecs)
132{
133 struct bio_vec *bv;
134 size_t orig_count = iov_iter_count(iter);
135 ssize_t bytes;
136 int npages;
137
138 iov_iter_truncate(iter, maxsize);
139 npages = iov_iter_npages(iter, INT_MAX);
140 iov_iter_reexpand(iter, orig_count);
141
142
143
144
145
146 bv = kvmalloc_array(npages, sizeof(*bv), GFP_KERNEL | __GFP_ZERO);
147 if (!bv)
148 return -ENOMEM;
149
150 bytes = __iter_get_bvecs(iter, maxsize, bv);
151 if (bytes < 0) {
152
153
154
155 kvfree(bv);
156 return bytes;
157 }
158
159 *bvecs = bv;
160 *num_bvecs = npages;
161 return bytes;
162}
163
164static void put_bvecs(struct bio_vec *bvecs, int num_bvecs, bool should_dirty)
165{
166 int i;
167
168 for (i = 0; i < num_bvecs; i++) {
169 if (bvecs[i].bv_page) {
170 if (should_dirty)
171 set_page_dirty_lock(bvecs[i].bv_page);
172 put_page(bvecs[i].bv_page);
173 }
174 }
175 kvfree(bvecs);
176}
177
178
179
180
181
182static struct ceph_mds_request *
183prepare_open_request(struct super_block *sb, int flags, int create_mode)
184{
185 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(sb);
186 struct ceph_mds_request *req;
187 int want_auth = USE_ANY_MDS;
188 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
189
190 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
191 want_auth = USE_AUTH_MDS;
192
193 req = ceph_mdsc_create_request(mdsc, op, want_auth);
194 if (IS_ERR(req))
195 goto out;
196 req->r_fmode = ceph_flags_to_mode(flags);
197 req->r_args.open.flags = ceph_flags_sys2wire(flags);
198 req->r_args.open.mode = cpu_to_le32(create_mode);
199out:
200 return req;
201}
202
203static int ceph_init_file_info(struct inode *inode, struct file *file,
204 int fmode, bool isdir)
205{
206 struct ceph_inode_info *ci = ceph_inode(inode);
207 struct ceph_file_info *fi;
208
209 dout("%s %p %p 0%o (%s)\n", __func__, inode, file,
210 inode->i_mode, isdir ? "dir" : "regular");
211 BUG_ON(inode->i_fop->release != ceph_release);
212
213 if (isdir) {
214 struct ceph_dir_file_info *dfi =
215 kmem_cache_zalloc(ceph_dir_file_cachep, GFP_KERNEL);
216 if (!dfi)
217 return -ENOMEM;
218
219 file->private_data = dfi;
220 fi = &dfi->file_info;
221 dfi->next_offset = 2;
222 dfi->readdir_cache_idx = -1;
223 } else {
224 fi = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL);
225 if (!fi)
226 return -ENOMEM;
227
228 file->private_data = fi;
229 }
230
231 ceph_get_fmode(ci, fmode, 1);
232 fi->fmode = fmode;
233
234 spin_lock_init(&fi->rw_contexts_lock);
235 INIT_LIST_HEAD(&fi->rw_contexts);
236 fi->meta_err = errseq_sample(&ci->i_meta_err);
237 fi->filp_gen = READ_ONCE(ceph_inode_to_client(inode)->filp_gen);
238
239 return 0;
240}
241
242
243
244
245
246static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
247{
248 int ret = 0;
249
250 switch (inode->i_mode & S_IFMT) {
251 case S_IFREG:
252 ceph_fscache_register_inode_cookie(inode);
253 ceph_fscache_file_set_cookie(inode, file);
254 fallthrough;
255 case S_IFDIR:
256 ret = ceph_init_file_info(inode, file, fmode,
257 S_ISDIR(inode->i_mode));
258 break;
259
260 case S_IFLNK:
261 dout("init_file %p %p 0%o (symlink)\n", inode, file,
262 inode->i_mode);
263 break;
264
265 default:
266 dout("init_file %p %p 0%o (special)\n", inode, file,
267 inode->i_mode);
268
269
270
271
272 BUG_ON(inode->i_fop->release == ceph_release);
273
274
275 ret = inode->i_fop->open(inode, file);
276 }
277 return ret;
278}
279
280
281
282
283int ceph_renew_caps(struct inode *inode, int fmode)
284{
285 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
286 struct ceph_inode_info *ci = ceph_inode(inode);
287 struct ceph_mds_request *req;
288 int err, flags, wanted;
289
290 spin_lock(&ci->i_ceph_lock);
291 __ceph_touch_fmode(ci, mdsc, fmode);
292 wanted = __ceph_caps_file_wanted(ci);
293 if (__ceph_is_any_real_caps(ci) &&
294 (!(wanted & CEPH_CAP_ANY_WR) || ci->i_auth_cap)) {
295 int issued = __ceph_caps_issued(ci, NULL);
296 spin_unlock(&ci->i_ceph_lock);
297 dout("renew caps %p want %s issued %s updating mds_wanted\n",
298 inode, ceph_cap_string(wanted), ceph_cap_string(issued));
299 ceph_check_caps(ci, 0, NULL);
300 return 0;
301 }
302 spin_unlock(&ci->i_ceph_lock);
303
304 flags = 0;
305 if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR))
306 flags = O_RDWR;
307 else if (wanted & CEPH_CAP_FILE_RD)
308 flags = O_RDONLY;
309 else if (wanted & CEPH_CAP_FILE_WR)
310 flags = O_WRONLY;
311#ifdef O_LAZY
312 if (wanted & CEPH_CAP_FILE_LAZYIO)
313 flags |= O_LAZY;
314#endif
315
316 req = prepare_open_request(inode->i_sb, flags, 0);
317 if (IS_ERR(req)) {
318 err = PTR_ERR(req);
319 goto out;
320 }
321
322 req->r_inode = inode;
323 ihold(inode);
324 req->r_num_caps = 1;
325
326 err = ceph_mdsc_do_request(mdsc, NULL, req);
327 ceph_mdsc_put_request(req);
328out:
329 dout("renew caps %p open result=%d\n", inode, err);
330 return err < 0 ? err : 0;
331}
332
333
334
335
336
337
338
339int ceph_open(struct inode *inode, struct file *file)
340{
341 struct ceph_inode_info *ci = ceph_inode(inode);
342 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
343 struct ceph_mds_client *mdsc = fsc->mdsc;
344 struct ceph_mds_request *req;
345 struct ceph_file_info *fi = file->private_data;
346 int err;
347 int flags, fmode, wanted;
348
349 if (fi) {
350 dout("open file %p is already opened\n", file);
351 return 0;
352 }
353
354
355 flags = file->f_flags & ~(O_CREAT|O_EXCL);
356 if (S_ISDIR(inode->i_mode))
357 flags = O_DIRECTORY;
358
359 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
360 ceph_vinop(inode), file, flags, file->f_flags);
361 fmode = ceph_flags_to_mode(flags);
362 wanted = ceph_caps_for_mode(fmode);
363
364
365 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
366 return -EROFS;
367
368
369 if (ceph_snap(inode) == CEPH_SNAPDIR) {
370 return ceph_init_file(inode, file, fmode);
371 }
372
373
374
375
376
377
378 spin_lock(&ci->i_ceph_lock);
379 if (__ceph_is_any_real_caps(ci) &&
380 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
381 int mds_wanted = __ceph_caps_mds_wanted(ci, true);
382 int issued = __ceph_caps_issued(ci, NULL);
383
384 dout("open %p fmode %d want %s issued %s using existing\n",
385 inode, fmode, ceph_cap_string(wanted),
386 ceph_cap_string(issued));
387 __ceph_touch_fmode(ci, mdsc, fmode);
388 spin_unlock(&ci->i_ceph_lock);
389
390
391 if ((issued & wanted) != wanted &&
392 (mds_wanted & wanted) != wanted &&
393 ceph_snap(inode) != CEPH_SNAPDIR)
394 ceph_check_caps(ci, 0, NULL);
395
396 return ceph_init_file(inode, file, fmode);
397 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
398 (ci->i_snap_caps & wanted) == wanted) {
399 __ceph_touch_fmode(ci, mdsc, fmode);
400 spin_unlock(&ci->i_ceph_lock);
401 return ceph_init_file(inode, file, fmode);
402 }
403
404 spin_unlock(&ci->i_ceph_lock);
405
406 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
407 req = prepare_open_request(inode->i_sb, flags, 0);
408 if (IS_ERR(req)) {
409 err = PTR_ERR(req);
410 goto out;
411 }
412 req->r_inode = inode;
413 ihold(inode);
414
415 req->r_num_caps = 1;
416 err = ceph_mdsc_do_request(mdsc, NULL, req);
417 if (!err)
418 err = ceph_init_file(inode, file, req->r_fmode);
419 ceph_mdsc_put_request(req);
420 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
421out:
422 return err;
423}
424
425
426static void
427cache_file_layout(struct inode *dst, struct inode *src)
428{
429 struct ceph_inode_info *cdst = ceph_inode(dst);
430 struct ceph_inode_info *csrc = ceph_inode(src);
431
432 spin_lock(&cdst->i_ceph_lock);
433 if ((__ceph_caps_issued(cdst, NULL) & CEPH_CAP_DIR_CREATE) &&
434 !ceph_file_layout_is_valid(&cdst->i_cached_layout)) {
435 memcpy(&cdst->i_cached_layout, &csrc->i_layout,
436 sizeof(cdst->i_cached_layout));
437 rcu_assign_pointer(cdst->i_cached_layout.pool_ns,
438 ceph_try_get_string(csrc->i_layout.pool_ns));
439 }
440 spin_unlock(&cdst->i_ceph_lock);
441}
442
443
444
445
446
447
448
449static int try_prep_async_create(struct inode *dir, struct dentry *dentry,
450 struct ceph_file_layout *lo, u64 *pino)
451{
452 struct ceph_inode_info *ci = ceph_inode(dir);
453 struct ceph_dentry_info *di = ceph_dentry(dentry);
454 int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_CREATE;
455 u64 ino;
456
457 spin_lock(&ci->i_ceph_lock);
458
459 if (!ci->i_auth_cap)
460 goto no_async;
461
462
463 if (xa_empty(&ci->i_auth_cap->session->s_delegated_inos))
464 goto no_async;
465
466 if (!ceph_file_layout_is_valid(&ci->i_cached_layout))
467 goto no_async;
468
469 if ((__ceph_caps_issued(ci, NULL) & want) != want)
470 goto no_async;
471
472 if (d_in_lookup(dentry)) {
473 if (!__ceph_dir_is_complete(ci))
474 goto no_async;
475 spin_lock(&dentry->d_lock);
476 di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
477 spin_unlock(&dentry->d_lock);
478 } else if (atomic_read(&ci->i_shared_gen) !=
479 READ_ONCE(di->lease_shared_gen)) {
480 goto no_async;
481 }
482
483 ino = ceph_get_deleg_ino(ci->i_auth_cap->session);
484 if (!ino)
485 goto no_async;
486
487 *pino = ino;
488 ceph_take_cap_refs(ci, want, false);
489 memcpy(lo, &ci->i_cached_layout, sizeof(*lo));
490 rcu_assign_pointer(lo->pool_ns,
491 ceph_try_get_string(ci->i_cached_layout.pool_ns));
492 got = want;
493no_async:
494 spin_unlock(&ci->i_ceph_lock);
495 return got;
496}
497
498static void restore_deleg_ino(struct inode *dir, u64 ino)
499{
500 struct ceph_inode_info *ci = ceph_inode(dir);
501 struct ceph_mds_session *s = NULL;
502
503 spin_lock(&ci->i_ceph_lock);
504 if (ci->i_auth_cap)
505 s = ceph_get_mds_session(ci->i_auth_cap->session);
506 spin_unlock(&ci->i_ceph_lock);
507 if (s) {
508 int err = ceph_restore_deleg_ino(s, ino);
509 if (err)
510 pr_warn("ceph: unable to restore delegated ino 0x%llx to session: %d\n",
511 ino, err);
512 ceph_put_mds_session(s);
513 }
514}
515
516static void ceph_async_create_cb(struct ceph_mds_client *mdsc,
517 struct ceph_mds_request *req)
518{
519 int result = req->r_err ? req->r_err :
520 le32_to_cpu(req->r_reply_info.head->result);
521
522 if (result == -EJUKEBOX)
523 goto out;
524
525 mapping_set_error(req->r_parent->i_mapping, result);
526
527 if (result) {
528 struct dentry *dentry = req->r_dentry;
529 int pathlen = 0;
530 u64 base = 0;
531 char *path = ceph_mdsc_build_path(req->r_dentry, &pathlen,
532 &base, 0);
533
534 ceph_dir_clear_complete(req->r_parent);
535 if (!d_unhashed(dentry))
536 d_drop(dentry);
537
538
539 pr_warn("ceph: async create failure path=(%llx)%s result=%d!\n",
540 base, IS_ERR(path) ? "<<bad>>" : path, result);
541 ceph_mdsc_free_path(path, pathlen);
542 }
543
544 if (req->r_target_inode) {
545 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
546 u64 ino = ceph_vino(req->r_target_inode).ino;
547
548 if (req->r_deleg_ino != ino)
549 pr_warn("%s: inode number mismatch! err=%d deleg_ino=0x%llx target=0x%llx\n",
550 __func__, req->r_err, req->r_deleg_ino, ino);
551 mapping_set_error(req->r_target_inode->i_mapping, result);
552
553 spin_lock(&ci->i_ceph_lock);
554 if (ci->i_ceph_flags & CEPH_I_ASYNC_CREATE) {
555 ci->i_ceph_flags &= ~CEPH_I_ASYNC_CREATE;
556 wake_up_bit(&ci->i_ceph_flags, CEPH_ASYNC_CREATE_BIT);
557 }
558 ceph_kick_flushing_inode_caps(req->r_session, ci);
559 spin_unlock(&ci->i_ceph_lock);
560 } else {
561 pr_warn("%s: no req->r_target_inode for 0x%llx\n", __func__,
562 req->r_deleg_ino);
563 }
564out:
565 ceph_mdsc_release_dir_caps(req);
566}
567
568static int ceph_finish_async_create(struct inode *dir, struct dentry *dentry,
569 struct file *file, umode_t mode,
570 struct ceph_mds_request *req,
571 struct ceph_acl_sec_ctx *as_ctx,
572 struct ceph_file_layout *lo)
573{
574 int ret;
575 char xattr_buf[4];
576 struct ceph_mds_reply_inode in = { };
577 struct ceph_mds_reply_info_in iinfo = { .in = &in };
578 struct ceph_inode_info *ci = ceph_inode(dir);
579 struct inode *inode;
580 struct timespec64 now;
581 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
582 struct ceph_vino vino = { .ino = req->r_deleg_ino,
583 .snap = CEPH_NOSNAP };
584
585 ktime_get_real_ts64(&now);
586
587 inode = ceph_get_inode(dentry->d_sb, vino);
588 if (IS_ERR(inode))
589 return PTR_ERR(inode);
590
591 iinfo.inline_version = CEPH_INLINE_NONE;
592 iinfo.change_attr = 1;
593 ceph_encode_timespec64(&iinfo.btime, &now);
594
595 iinfo.xattr_len = ARRAY_SIZE(xattr_buf);
596 iinfo.xattr_data = xattr_buf;
597 memset(iinfo.xattr_data, 0, iinfo.xattr_len);
598
599 in.ino = cpu_to_le64(vino.ino);
600 in.snapid = cpu_to_le64(CEPH_NOSNAP);
601 in.version = cpu_to_le64(1);
602 in.cap.caps = in.cap.wanted = cpu_to_le32(CEPH_CAP_ALL_FILE);
603 in.cap.cap_id = cpu_to_le64(1);
604 in.cap.realm = cpu_to_le64(ci->i_snap_realm->ino);
605 in.cap.flags = CEPH_CAP_FLAG_AUTH;
606 in.ctime = in.mtime = in.atime = iinfo.btime;
607 in.mode = cpu_to_le32((u32)mode);
608 in.truncate_seq = cpu_to_le32(1);
609 in.truncate_size = cpu_to_le64(-1ULL);
610 in.xattr_version = cpu_to_le64(1);
611 in.uid = cpu_to_le32(from_kuid(&init_user_ns, current_fsuid()));
612 in.gid = cpu_to_le32(from_kgid(&init_user_ns, dir->i_mode & S_ISGID ?
613 dir->i_gid : current_fsgid()));
614 in.nlink = cpu_to_le32(1);
615 in.max_size = cpu_to_le64(lo->stripe_unit);
616
617 ceph_file_layout_to_legacy(lo, &in.layout);
618
619 down_read(&mdsc->snap_rwsem);
620 ret = ceph_fill_inode(inode, NULL, &iinfo, NULL, req->r_session,
621 req->r_fmode, NULL);
622 up_read(&mdsc->snap_rwsem);
623 if (ret) {
624 dout("%s failed to fill inode: %d\n", __func__, ret);
625 ceph_dir_clear_complete(dir);
626 if (!d_unhashed(dentry))
627 d_drop(dentry);
628 if (inode->i_state & I_NEW)
629 discard_new_inode(inode);
630 } else {
631 struct dentry *dn;
632
633 dout("%s d_adding new inode 0x%llx to 0x%llx/%s\n", __func__,
634 vino.ino, ceph_ino(dir), dentry->d_name.name);
635 ceph_dir_clear_ordered(dir);
636 ceph_init_inode_acls(inode, as_ctx);
637 if (inode->i_state & I_NEW) {
638
639
640
641
642
643
644 ceph_inode(inode)->i_ceph_flags = CEPH_I_ASYNC_CREATE;
645 unlock_new_inode(inode);
646 }
647 if (d_in_lookup(dentry) || d_really_is_negative(dentry)) {
648 if (!d_unhashed(dentry))
649 d_drop(dentry);
650 dn = d_splice_alias(inode, dentry);
651 WARN_ON_ONCE(dn && dn != dentry);
652 }
653 file->f_mode |= FMODE_CREATED;
654 ret = finish_open(file, dentry, ceph_open);
655 }
656 return ret;
657}
658
659
660
661
662
663int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
664 struct file *file, unsigned flags, umode_t mode)
665{
666 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
667 struct ceph_mds_client *mdsc = fsc->mdsc;
668 struct ceph_mds_request *req;
669 struct dentry *dn;
670 struct ceph_acl_sec_ctx as_ctx = {};
671 bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS);
672 int mask;
673 int err;
674
675 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
676 dir, dentry, dentry,
677 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
678
679 if (dentry->d_name.len > NAME_MAX)
680 return -ENAMETOOLONG;
681
682 if (flags & O_CREAT) {
683 if (ceph_quota_is_max_files_exceeded(dir))
684 return -EDQUOT;
685 err = ceph_pre_init_acls(dir, &mode, &as_ctx);
686 if (err < 0)
687 return err;
688 err = ceph_security_init_secctx(dentry, mode, &as_ctx);
689 if (err < 0)
690 goto out_ctx;
691 } else if (!d_in_lookup(dentry)) {
692
693 return -ENOENT;
694 }
695retry:
696
697 req = prepare_open_request(dir->i_sb, flags, mode);
698 if (IS_ERR(req)) {
699 err = PTR_ERR(req);
700 goto out_ctx;
701 }
702 req->r_dentry = dget(dentry);
703 req->r_num_caps = 2;
704 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
705 if (ceph_security_xattr_wanted(dir))
706 mask |= CEPH_CAP_XATTR_SHARED;
707 req->r_args.open.mask = cpu_to_le32(mask);
708 req->r_parent = dir;
709 ihold(dir);
710
711 if (flags & O_CREAT) {
712 struct ceph_file_layout lo;
713
714 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
715 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
716 if (as_ctx.pagelist) {
717 req->r_pagelist = as_ctx.pagelist;
718 as_ctx.pagelist = NULL;
719 }
720 if (try_async &&
721 (req->r_dir_caps =
722 try_prep_async_create(dir, dentry, &lo,
723 &req->r_deleg_ino))) {
724 set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags);
725 req->r_args.open.flags |= cpu_to_le32(CEPH_O_EXCL);
726 req->r_callback = ceph_async_create_cb;
727 err = ceph_mdsc_submit_request(mdsc, dir, req);
728 if (!err) {
729 err = ceph_finish_async_create(dir, dentry,
730 file, mode, req,
731 &as_ctx, &lo);
732 } else if (err == -EJUKEBOX) {
733 restore_deleg_ino(dir, req->r_deleg_ino);
734 ceph_mdsc_put_request(req);
735 try_async = false;
736 goto retry;
737 }
738 goto out_req;
739 }
740 }
741
742 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
743 err = ceph_mdsc_do_request(mdsc,
744 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
745 req);
746 if (err == -ENOENT) {
747 dentry = ceph_handle_snapdir(req, dentry);
748 if (IS_ERR(dentry)) {
749 err = PTR_ERR(dentry);
750 goto out_req;
751 }
752 err = 0;
753 }
754
755 if (!err && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
756 err = ceph_handle_notrace_create(dir, dentry);
757
758 if (d_in_lookup(dentry)) {
759 dn = ceph_finish_lookup(req, dentry, err);
760 if (IS_ERR(dn))
761 err = PTR_ERR(dn);
762 } else {
763
764 dn = NULL;
765 }
766 if (err)
767 goto out_req;
768 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
769
770 dout("atomic_open finish_no_open on dn %p\n", dn);
771 err = finish_no_open(file, dn);
772 } else {
773 dout("atomic_open finish_open on dn %p\n", dn);
774 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
775 struct inode *newino = d_inode(dentry);
776
777 cache_file_layout(dir, newino);
778 ceph_init_inode_acls(newino, &as_ctx);
779 file->f_mode |= FMODE_CREATED;
780 }
781 err = finish_open(file, dentry, ceph_open);
782 }
783out_req:
784 ceph_mdsc_put_request(req);
785out_ctx:
786 ceph_release_acl_sec_ctx(&as_ctx);
787 dout("atomic_open result=%d\n", err);
788 return err;
789}
790
791int ceph_release(struct inode *inode, struct file *file)
792{
793 struct ceph_inode_info *ci = ceph_inode(inode);
794
795 if (S_ISDIR(inode->i_mode)) {
796 struct ceph_dir_file_info *dfi = file->private_data;
797 dout("release inode %p dir file %p\n", inode, file);
798 WARN_ON(!list_empty(&dfi->file_info.rw_contexts));
799
800 ceph_put_fmode(ci, dfi->file_info.fmode, 1);
801
802 if (dfi->last_readdir)
803 ceph_mdsc_put_request(dfi->last_readdir);
804 kfree(dfi->last_name);
805 kfree(dfi->dir_info);
806 kmem_cache_free(ceph_dir_file_cachep, dfi);
807 } else {
808 struct ceph_file_info *fi = file->private_data;
809 dout("release inode %p regular file %p\n", inode, file);
810 WARN_ON(!list_empty(&fi->rw_contexts));
811
812 ceph_put_fmode(ci, fi->fmode, 1);
813
814 kmem_cache_free(ceph_file_cachep, fi);
815 }
816
817
818 wake_up_all(&ci->i_cap_wq);
819 return 0;
820}
821
822enum {
823 HAVE_RETRIED = 1,
824 CHECK_EOF = 2,
825 READ_INLINE = 3,
826};
827
828
829
830
831
832
833
834
835
836
837
838static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
839 int *retry_op)
840{
841 struct file *file = iocb->ki_filp;
842 struct inode *inode = file_inode(file);
843 struct ceph_inode_info *ci = ceph_inode(inode);
844 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
845 struct ceph_osd_client *osdc = &fsc->client->osdc;
846 ssize_t ret;
847 u64 off = iocb->ki_pos;
848 u64 len = iov_iter_count(to);
849
850 dout("sync_read on file %p %llu~%u %s\n", file, off, (unsigned)len,
851 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
852
853 if (!len)
854 return 0;
855
856
857
858
859
860
861 ret = filemap_write_and_wait_range(inode->i_mapping,
862 off, off + len - 1);
863 if (ret < 0)
864 return ret;
865
866 ret = 0;
867 while ((len = iov_iter_count(to)) > 0) {
868 struct ceph_osd_request *req;
869 struct page **pages;
870 int num_pages;
871 size_t page_off;
872 u64 i_size;
873 bool more;
874 int idx;
875 size_t left;
876
877 req = ceph_osdc_new_request(osdc, &ci->i_layout,
878 ci->i_vino, off, &len, 0, 1,
879 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
880 NULL, ci->i_truncate_seq,
881 ci->i_truncate_size, false);
882 if (IS_ERR(req)) {
883 ret = PTR_ERR(req);
884 break;
885 }
886
887 more = len < iov_iter_count(to);
888
889 num_pages = calc_pages_for(off, len);
890 page_off = off & ~PAGE_MASK;
891 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
892 if (IS_ERR(pages)) {
893 ceph_osdc_put_request(req);
894 ret = PTR_ERR(pages);
895 break;
896 }
897
898 osd_req_op_extent_osd_data_pages(req, 0, pages, len, page_off,
899 false, false);
900 ret = ceph_osdc_start_request(osdc, req, false);
901 if (!ret)
902 ret = ceph_osdc_wait_request(osdc, req);
903
904 ceph_update_read_metrics(&fsc->mdsc->metric,
905 req->r_start_latency,
906 req->r_end_latency,
907 len, ret);
908
909 ceph_osdc_put_request(req);
910
911 i_size = i_size_read(inode);
912 dout("sync_read %llu~%llu got %zd i_size %llu%s\n",
913 off, len, ret, i_size, (more ? " MORE" : ""));
914
915 if (ret == -ENOENT)
916 ret = 0;
917 if (ret >= 0 && ret < len && (off + ret < i_size)) {
918 int zlen = min(len - ret, i_size - off - ret);
919 int zoff = page_off + ret;
920 dout("sync_read zero gap %llu~%llu\n",
921 off + ret, off + ret + zlen);
922 ceph_zero_page_vector_range(zoff, zlen, pages);
923 ret += zlen;
924 }
925
926 idx = 0;
927 left = ret > 0 ? ret : 0;
928 while (left > 0) {
929 size_t len, copied;
930 page_off = off & ~PAGE_MASK;
931 len = min_t(size_t, left, PAGE_SIZE - page_off);
932 SetPageUptodate(pages[idx]);
933 copied = copy_page_to_iter(pages[idx++],
934 page_off, len, to);
935 off += copied;
936 left -= copied;
937 if (copied < len) {
938 ret = -EFAULT;
939 break;
940 }
941 }
942 ceph_release_page_vector(pages, num_pages);
943
944 if (ret < 0) {
945 if (ret == -EBLOCKLISTED)
946 fsc->blocklisted = true;
947 break;
948 }
949
950 if (off >= i_size || !more)
951 break;
952 }
953
954 if (off > iocb->ki_pos) {
955 if (ret >= 0 &&
956 iov_iter_count(to) > 0 && off >= i_size_read(inode))
957 *retry_op = CHECK_EOF;
958 ret = off - iocb->ki_pos;
959 iocb->ki_pos = off;
960 }
961
962 dout("sync_read result %zd retry_op %d\n", ret, *retry_op);
963 return ret;
964}
965
966struct ceph_aio_request {
967 struct kiocb *iocb;
968 size_t total_len;
969 bool write;
970 bool should_dirty;
971 int error;
972 struct list_head osd_reqs;
973 unsigned num_reqs;
974 atomic_t pending_reqs;
975 struct timespec64 mtime;
976 struct ceph_cap_flush *prealloc_cf;
977};
978
979struct ceph_aio_work {
980 struct work_struct work;
981 struct ceph_osd_request *req;
982};
983
984static void ceph_aio_retry_work(struct work_struct *work);
985
986static void ceph_aio_complete(struct inode *inode,
987 struct ceph_aio_request *aio_req)
988{
989 struct ceph_inode_info *ci = ceph_inode(inode);
990 int ret;
991
992 if (!atomic_dec_and_test(&aio_req->pending_reqs))
993 return;
994
995 if (aio_req->iocb->ki_flags & IOCB_DIRECT)
996 inode_dio_end(inode);
997
998 ret = aio_req->error;
999 if (!ret)
1000 ret = aio_req->total_len;
1001
1002 dout("ceph_aio_complete %p rc %d\n", inode, ret);
1003
1004 if (ret >= 0 && aio_req->write) {
1005 int dirty;
1006
1007 loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len;
1008 if (endoff > i_size_read(inode)) {
1009 if (ceph_inode_set_size(inode, endoff))
1010 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1011 }
1012
1013 spin_lock(&ci->i_ceph_lock);
1014 ci->i_inline_version = CEPH_INLINE_NONE;
1015 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1016 &aio_req->prealloc_cf);
1017 spin_unlock(&ci->i_ceph_lock);
1018 if (dirty)
1019 __mark_inode_dirty(inode, dirty);
1020
1021 }
1022
1023 ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR :
1024 CEPH_CAP_FILE_RD));
1025
1026 aio_req->iocb->ki_complete(aio_req->iocb, ret, 0);
1027
1028 ceph_free_cap_flush(aio_req->prealloc_cf);
1029 kfree(aio_req);
1030}
1031
1032static void ceph_aio_complete_req(struct ceph_osd_request *req)
1033{
1034 int rc = req->r_result;
1035 struct inode *inode = req->r_inode;
1036 struct ceph_aio_request *aio_req = req->r_priv;
1037 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
1038 struct ceph_client_metric *metric = &ceph_sb_to_mdsc(inode->i_sb)->metric;
1039 unsigned int len = osd_data->bvec_pos.iter.bi_size;
1040
1041 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_BVECS);
1042 BUG_ON(!osd_data->num_bvecs);
1043
1044 dout("ceph_aio_complete_req %p rc %d bytes %u\n", inode, rc, len);
1045
1046 if (rc == -EOLDSNAPC) {
1047 struct ceph_aio_work *aio_work;
1048 BUG_ON(!aio_req->write);
1049
1050 aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS);
1051 if (aio_work) {
1052 INIT_WORK(&aio_work->work, ceph_aio_retry_work);
1053 aio_work->req = req;
1054 queue_work(ceph_inode_to_client(inode)->inode_wq,
1055 &aio_work->work);
1056 return;
1057 }
1058 rc = -ENOMEM;
1059 } else if (!aio_req->write) {
1060 if (rc == -ENOENT)
1061 rc = 0;
1062 if (rc >= 0 && len > rc) {
1063 struct iov_iter i;
1064 int zlen = len - rc;
1065
1066
1067
1068
1069
1070
1071 if (aio_req->num_reqs == 1) {
1072 loff_t i_size = i_size_read(inode);
1073 loff_t endoff = aio_req->iocb->ki_pos + rc;
1074 if (endoff < i_size)
1075 zlen = min_t(size_t, zlen,
1076 i_size - endoff);
1077 aio_req->total_len = rc + zlen;
1078 }
1079
1080 iov_iter_bvec(&i, READ, osd_data->bvec_pos.bvecs,
1081 osd_data->num_bvecs, len);
1082 iov_iter_advance(&i, rc);
1083 iov_iter_zero(zlen, &i);
1084 }
1085 }
1086
1087
1088 if (req->r_start_latency) {
1089 if (aio_req->write)
1090 ceph_update_write_metrics(metric, req->r_start_latency,
1091 req->r_end_latency, len, rc);
1092 else
1093 ceph_update_read_metrics(metric, req->r_start_latency,
1094 req->r_end_latency, len, rc);
1095 }
1096
1097 put_bvecs(osd_data->bvec_pos.bvecs, osd_data->num_bvecs,
1098 aio_req->should_dirty);
1099 ceph_osdc_put_request(req);
1100
1101 if (rc < 0)
1102 cmpxchg(&aio_req->error, 0, rc);
1103
1104 ceph_aio_complete(inode, aio_req);
1105 return;
1106}
1107
1108static void ceph_aio_retry_work(struct work_struct *work)
1109{
1110 struct ceph_aio_work *aio_work =
1111 container_of(work, struct ceph_aio_work, work);
1112 struct ceph_osd_request *orig_req = aio_work->req;
1113 struct ceph_aio_request *aio_req = orig_req->r_priv;
1114 struct inode *inode = orig_req->r_inode;
1115 struct ceph_inode_info *ci = ceph_inode(inode);
1116 struct ceph_snap_context *snapc;
1117 struct ceph_osd_request *req;
1118 int ret;
1119
1120 spin_lock(&ci->i_ceph_lock);
1121 if (__ceph_have_pending_cap_snap(ci)) {
1122 struct ceph_cap_snap *capsnap =
1123 list_last_entry(&ci->i_cap_snaps,
1124 struct ceph_cap_snap,
1125 ci_item);
1126 snapc = ceph_get_snap_context(capsnap->context);
1127 } else {
1128 BUG_ON(!ci->i_head_snapc);
1129 snapc = ceph_get_snap_context(ci->i_head_snapc);
1130 }
1131 spin_unlock(&ci->i_ceph_lock);
1132
1133 req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 1,
1134 false, GFP_NOFS);
1135 if (!req) {
1136 ret = -ENOMEM;
1137 req = orig_req;
1138 goto out;
1139 }
1140
1141 req->r_flags = CEPH_OSD_FLAG_WRITE;
1142 ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc);
1143 ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid);
1144
1145 req->r_ops[0] = orig_req->r_ops[0];
1146
1147 req->r_mtime = aio_req->mtime;
1148 req->r_data_offset = req->r_ops[0].extent.offset;
1149
1150 ret = ceph_osdc_alloc_messages(req, GFP_NOFS);
1151 if (ret) {
1152 ceph_osdc_put_request(req);
1153 req = orig_req;
1154 goto out;
1155 }
1156
1157 ceph_osdc_put_request(orig_req);
1158
1159 req->r_callback = ceph_aio_complete_req;
1160 req->r_inode = inode;
1161 req->r_priv = aio_req;
1162
1163 ret = ceph_osdc_start_request(req->r_osdc, req, false);
1164out:
1165 if (ret < 0) {
1166 req->r_result = ret;
1167 ceph_aio_complete_req(req);
1168 }
1169
1170 ceph_put_snap_context(snapc);
1171 kfree(aio_work);
1172}
1173
1174static ssize_t
1175ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
1176 struct ceph_snap_context *snapc,
1177 struct ceph_cap_flush **pcf)
1178{
1179 struct file *file = iocb->ki_filp;
1180 struct inode *inode = file_inode(file);
1181 struct ceph_inode_info *ci = ceph_inode(inode);
1182 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1183 struct ceph_client_metric *metric = &fsc->mdsc->metric;
1184 struct ceph_vino vino;
1185 struct ceph_osd_request *req;
1186 struct bio_vec *bvecs;
1187 struct ceph_aio_request *aio_req = NULL;
1188 int num_pages = 0;
1189 int flags;
1190 int ret = 0;
1191 struct timespec64 mtime = current_time(inode);
1192 size_t count = iov_iter_count(iter);
1193 loff_t pos = iocb->ki_pos;
1194 bool write = iov_iter_rw(iter) == WRITE;
1195 bool should_dirty = !write && iter_is_iovec(iter);
1196
1197 if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1198 return -EROFS;
1199
1200 dout("sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n",
1201 (write ? "write" : "read"), file, pos, (unsigned)count,
1202 snapc, snapc ? snapc->seq : 0);
1203
1204 if (write) {
1205 int ret2 = invalidate_inode_pages2_range(inode->i_mapping,
1206 pos >> PAGE_SHIFT,
1207 (pos + count - 1) >> PAGE_SHIFT);
1208 if (ret2 < 0)
1209 dout("invalidate_inode_pages2_range returned %d\n", ret2);
1210
1211 flags = CEPH_OSD_FLAG_WRITE;
1212 } else {
1213 flags = CEPH_OSD_FLAG_READ;
1214 }
1215
1216 while (iov_iter_count(iter) > 0) {
1217 u64 size = iov_iter_count(iter);
1218 ssize_t len;
1219
1220 if (write)
1221 size = min_t(u64, size, fsc->mount_options->wsize);
1222 else
1223 size = min_t(u64, size, fsc->mount_options->rsize);
1224
1225 vino = ceph_vino(inode);
1226 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1227 vino, pos, &size, 0,
1228 1,
1229 write ? CEPH_OSD_OP_WRITE :
1230 CEPH_OSD_OP_READ,
1231 flags, snapc,
1232 ci->i_truncate_seq,
1233 ci->i_truncate_size,
1234 false);
1235 if (IS_ERR(req)) {
1236 ret = PTR_ERR(req);
1237 break;
1238 }
1239
1240 len = iter_get_bvecs_alloc(iter, size, &bvecs, &num_pages);
1241 if (len < 0) {
1242 ceph_osdc_put_request(req);
1243 ret = len;
1244 break;
1245 }
1246 if (len != size)
1247 osd_req_op_extent_update(req, 0, len);
1248
1249
1250
1251
1252
1253 if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) &&
1254 (len == count || pos + count <= i_size_read(inode))) {
1255 aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL);
1256 if (aio_req) {
1257 aio_req->iocb = iocb;
1258 aio_req->write = write;
1259 aio_req->should_dirty = should_dirty;
1260 INIT_LIST_HEAD(&aio_req->osd_reqs);
1261 if (write) {
1262 aio_req->mtime = mtime;
1263 swap(aio_req->prealloc_cf, *pcf);
1264 }
1265 }
1266
1267 }
1268
1269 if (write) {
1270
1271
1272
1273
1274 truncate_inode_pages_range(inode->i_mapping, pos,
1275 PAGE_ALIGN(pos + len) - 1);
1276
1277 req->r_mtime = mtime;
1278 }
1279
1280 osd_req_op_extent_osd_data_bvecs(req, 0, bvecs, num_pages, len);
1281
1282 if (aio_req) {
1283 aio_req->total_len += len;
1284 aio_req->num_reqs++;
1285 atomic_inc(&aio_req->pending_reqs);
1286
1287 req->r_callback = ceph_aio_complete_req;
1288 req->r_inode = inode;
1289 req->r_priv = aio_req;
1290 list_add_tail(&req->r_private_item, &aio_req->osd_reqs);
1291
1292 pos += len;
1293 continue;
1294 }
1295
1296 ret = ceph_osdc_start_request(req->r_osdc, req, false);
1297 if (!ret)
1298 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1299
1300 if (write)
1301 ceph_update_write_metrics(metric, req->r_start_latency,
1302 req->r_end_latency, len, ret);
1303 else
1304 ceph_update_read_metrics(metric, req->r_start_latency,
1305 req->r_end_latency, len, ret);
1306
1307 size = i_size_read(inode);
1308 if (!write) {
1309 if (ret == -ENOENT)
1310 ret = 0;
1311 if (ret >= 0 && ret < len && pos + ret < size) {
1312 struct iov_iter i;
1313 int zlen = min_t(size_t, len - ret,
1314 size - pos - ret);
1315
1316 iov_iter_bvec(&i, READ, bvecs, num_pages, len);
1317 iov_iter_advance(&i, ret);
1318 iov_iter_zero(zlen, &i);
1319 ret += zlen;
1320 }
1321 if (ret >= 0)
1322 len = ret;
1323 }
1324
1325 put_bvecs(bvecs, num_pages, should_dirty);
1326 ceph_osdc_put_request(req);
1327 if (ret < 0)
1328 break;
1329
1330 pos += len;
1331 if (!write && pos >= size)
1332 break;
1333
1334 if (write && pos > size) {
1335 if (ceph_inode_set_size(inode, pos))
1336 ceph_check_caps(ceph_inode(inode),
1337 CHECK_CAPS_AUTHONLY,
1338 NULL);
1339 }
1340 }
1341
1342 if (aio_req) {
1343 LIST_HEAD(osd_reqs);
1344
1345 if (aio_req->num_reqs == 0) {
1346 kfree(aio_req);
1347 return ret;
1348 }
1349
1350 ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR :
1351 CEPH_CAP_FILE_RD);
1352
1353 list_splice(&aio_req->osd_reqs, &osd_reqs);
1354 inode_dio_begin(inode);
1355 while (!list_empty(&osd_reqs)) {
1356 req = list_first_entry(&osd_reqs,
1357 struct ceph_osd_request,
1358 r_private_item);
1359 list_del_init(&req->r_private_item);
1360 if (ret >= 0)
1361 ret = ceph_osdc_start_request(req->r_osdc,
1362 req, false);
1363 if (ret < 0) {
1364 req->r_result = ret;
1365 ceph_aio_complete_req(req);
1366 }
1367 }
1368 return -EIOCBQUEUED;
1369 }
1370
1371 if (ret != -EOLDSNAPC && pos > iocb->ki_pos) {
1372 ret = pos - iocb->ki_pos;
1373 iocb->ki_pos = pos;
1374 }
1375 return ret;
1376}
1377
1378
1379
1380
1381
1382
1383
1384
1385static ssize_t
1386ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
1387 struct ceph_snap_context *snapc)
1388{
1389 struct file *file = iocb->ki_filp;
1390 struct inode *inode = file_inode(file);
1391 struct ceph_inode_info *ci = ceph_inode(inode);
1392 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1393 struct ceph_vino vino;
1394 struct ceph_osd_request *req;
1395 struct page **pages;
1396 u64 len;
1397 int num_pages;
1398 int written = 0;
1399 int flags;
1400 int ret;
1401 bool check_caps = false;
1402 struct timespec64 mtime = current_time(inode);
1403 size_t count = iov_iter_count(from);
1404
1405 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
1406 return -EROFS;
1407
1408 dout("sync_write on file %p %lld~%u snapc %p seq %lld\n",
1409 file, pos, (unsigned)count, snapc, snapc->seq);
1410
1411 ret = filemap_write_and_wait_range(inode->i_mapping,
1412 pos, pos + count - 1);
1413 if (ret < 0)
1414 return ret;
1415
1416 ret = invalidate_inode_pages2_range(inode->i_mapping,
1417 pos >> PAGE_SHIFT,
1418 (pos + count - 1) >> PAGE_SHIFT);
1419 if (ret < 0)
1420 dout("invalidate_inode_pages2_range returned %d\n", ret);
1421
1422 flags = CEPH_OSD_FLAG_WRITE;
1423
1424 while ((len = iov_iter_count(from)) > 0) {
1425 size_t left;
1426 int n;
1427
1428 vino = ceph_vino(inode);
1429 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1430 vino, pos, &len, 0, 1,
1431 CEPH_OSD_OP_WRITE, flags, snapc,
1432 ci->i_truncate_seq,
1433 ci->i_truncate_size,
1434 false);
1435 if (IS_ERR(req)) {
1436 ret = PTR_ERR(req);
1437 break;
1438 }
1439
1440
1441
1442
1443
1444 num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1445
1446 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
1447 if (IS_ERR(pages)) {
1448 ret = PTR_ERR(pages);
1449 goto out;
1450 }
1451
1452 left = len;
1453 for (n = 0; n < num_pages; n++) {
1454 size_t plen = min_t(size_t, left, PAGE_SIZE);
1455 ret = copy_page_from_iter(pages[n], 0, plen, from);
1456 if (ret != plen) {
1457 ret = -EFAULT;
1458 break;
1459 }
1460 left -= ret;
1461 }
1462
1463 if (ret < 0) {
1464 ceph_release_page_vector(pages, num_pages);
1465 goto out;
1466 }
1467
1468 req->r_inode = inode;
1469
1470 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
1471 false, true);
1472
1473 req->r_mtime = mtime;
1474 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1475 if (!ret)
1476 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1477
1478 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
1479 req->r_end_latency, len, ret);
1480out:
1481 ceph_osdc_put_request(req);
1482 if (ret != 0) {
1483 ceph_set_error_write(ci);
1484 break;
1485 }
1486
1487 ceph_clear_error_write(ci);
1488 pos += len;
1489 written += len;
1490 if (pos > i_size_read(inode)) {
1491 check_caps = ceph_inode_set_size(inode, pos);
1492 if (check_caps)
1493 ceph_check_caps(ceph_inode(inode),
1494 CHECK_CAPS_AUTHONLY,
1495 NULL);
1496 }
1497
1498 }
1499
1500 if (ret != -EOLDSNAPC && written > 0) {
1501 ret = written;
1502 iocb->ki_pos = pos;
1503 }
1504 return ret;
1505}
1506
1507
1508
1509
1510
1511
1512
1513
1514static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
1515{
1516 struct file *filp = iocb->ki_filp;
1517 struct ceph_file_info *fi = filp->private_data;
1518 size_t len = iov_iter_count(to);
1519 struct inode *inode = file_inode(filp);
1520 struct ceph_inode_info *ci = ceph_inode(inode);
1521 bool direct_lock = iocb->ki_flags & IOCB_DIRECT;
1522 ssize_t ret;
1523 int want, got = 0;
1524 int retry_op = 0, read = 0;
1525
1526again:
1527 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
1528 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
1529
1530 if (direct_lock)
1531 ceph_start_io_direct(inode);
1532 else
1533 ceph_start_io_read(inode);
1534
1535 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1536 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1537 else
1538 want = CEPH_CAP_FILE_CACHE;
1539 ret = ceph_get_caps(filp, CEPH_CAP_FILE_RD, want, -1, &got);
1540 if (ret < 0) {
1541 if (iocb->ki_flags & IOCB_DIRECT)
1542 ceph_end_io_direct(inode);
1543 else
1544 ceph_end_io_read(inode);
1545 return ret;
1546 }
1547
1548 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1549 (iocb->ki_flags & IOCB_DIRECT) ||
1550 (fi->flags & CEPH_F_SYNC)) {
1551
1552 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1553 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1554 ceph_cap_string(got));
1555
1556 if (ci->i_inline_version == CEPH_INLINE_NONE) {
1557 if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) {
1558 ret = ceph_direct_read_write(iocb, to,
1559 NULL, NULL);
1560 if (ret >= 0 && ret < len)
1561 retry_op = CHECK_EOF;
1562 } else {
1563 ret = ceph_sync_read(iocb, to, &retry_op);
1564 }
1565 } else {
1566 retry_op = READ_INLINE;
1567 }
1568 } else {
1569 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1570 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
1571 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
1572 ceph_cap_string(got));
1573 ceph_add_rw_context(fi, &rw_ctx);
1574 ret = generic_file_read_iter(iocb, to);
1575 ceph_del_rw_context(fi, &rw_ctx);
1576 }
1577
1578 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
1579 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
1580 ceph_put_cap_refs(ci, got);
1581
1582 if (direct_lock)
1583 ceph_end_io_direct(inode);
1584 else
1585 ceph_end_io_read(inode);
1586
1587 if (retry_op > HAVE_RETRIED && ret >= 0) {
1588 int statret;
1589 struct page *page = NULL;
1590 loff_t i_size;
1591 if (retry_op == READ_INLINE) {
1592 page = __page_cache_alloc(GFP_KERNEL);
1593 if (!page)
1594 return -ENOMEM;
1595 }
1596
1597 statret = __ceph_do_getattr(inode, page,
1598 CEPH_STAT_CAP_INLINE_DATA, !!page);
1599 if (statret < 0) {
1600 if (page)
1601 __free_page(page);
1602 if (statret == -ENODATA) {
1603 BUG_ON(retry_op != READ_INLINE);
1604 goto again;
1605 }
1606 return statret;
1607 }
1608
1609 i_size = i_size_read(inode);
1610 if (retry_op == READ_INLINE) {
1611 BUG_ON(ret > 0 || read > 0);
1612 if (iocb->ki_pos < i_size &&
1613 iocb->ki_pos < PAGE_SIZE) {
1614 loff_t end = min_t(loff_t, i_size,
1615 iocb->ki_pos + len);
1616 end = min_t(loff_t, end, PAGE_SIZE);
1617 if (statret < end)
1618 zero_user_segment(page, statret, end);
1619 ret = copy_page_to_iter(page,
1620 iocb->ki_pos & ~PAGE_MASK,
1621 end - iocb->ki_pos, to);
1622 iocb->ki_pos += ret;
1623 read += ret;
1624 }
1625 if (iocb->ki_pos < i_size && read < len) {
1626 size_t zlen = min_t(size_t, len - read,
1627 i_size - iocb->ki_pos);
1628 ret = iov_iter_zero(zlen, to);
1629 iocb->ki_pos += ret;
1630 read += ret;
1631 }
1632 __free_pages(page, 0);
1633 return read;
1634 }
1635
1636
1637 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
1638 ret < len) {
1639 dout("sync_read hit hole, ppos %lld < size %lld"
1640 ", reading more\n", iocb->ki_pos, i_size);
1641
1642 read += ret;
1643 len -= ret;
1644 retry_op = HAVE_RETRIED;
1645 goto again;
1646 }
1647 }
1648
1649 if (ret >= 0)
1650 ret += read;
1651
1652 return ret;
1653}
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
1666{
1667 struct file *file = iocb->ki_filp;
1668 struct ceph_file_info *fi = file->private_data;
1669 struct inode *inode = file_inode(file);
1670 struct ceph_inode_info *ci = ceph_inode(inode);
1671 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1672 struct ceph_osd_client *osdc = &fsc->client->osdc;
1673 struct ceph_cap_flush *prealloc_cf;
1674 ssize_t count, written = 0;
1675 int err, want, got;
1676 bool direct_lock = false;
1677 u32 map_flags;
1678 u64 pool_flags;
1679 loff_t pos;
1680 loff_t limit = max(i_size_read(inode), fsc->max_file_size);
1681
1682 if (ceph_snap(inode) != CEPH_NOSNAP)
1683 return -EROFS;
1684
1685 prealloc_cf = ceph_alloc_cap_flush();
1686 if (!prealloc_cf)
1687 return -ENOMEM;
1688
1689 if ((iocb->ki_flags & (IOCB_DIRECT | IOCB_APPEND)) == IOCB_DIRECT)
1690 direct_lock = true;
1691
1692retry_snap:
1693 if (direct_lock)
1694 ceph_start_io_direct(inode);
1695 else
1696 ceph_start_io_write(inode);
1697
1698
1699 current->backing_dev_info = inode_to_bdi(inode);
1700
1701 if (iocb->ki_flags & IOCB_APPEND) {
1702 err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1703 if (err < 0)
1704 goto out;
1705 }
1706
1707 err = generic_write_checks(iocb, from);
1708 if (err <= 0)
1709 goto out;
1710
1711 pos = iocb->ki_pos;
1712 if (unlikely(pos >= limit)) {
1713 err = -EFBIG;
1714 goto out;
1715 } else {
1716 iov_iter_truncate(from, limit - pos);
1717 }
1718
1719 count = iov_iter_count(from);
1720 if (ceph_quota_is_max_bytes_exceeded(inode, pos + count)) {
1721 err = -EDQUOT;
1722 goto out;
1723 }
1724
1725 err = file_remove_privs(file);
1726 if (err)
1727 goto out;
1728
1729 err = file_update_time(file);
1730 if (err)
1731 goto out;
1732
1733 inode_inc_iversion_raw(inode);
1734
1735 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1736 err = ceph_uninline_data(file, NULL);
1737 if (err < 0)
1738 goto out;
1739 }
1740
1741 down_read(&osdc->lock);
1742 map_flags = osdc->osdmap->flags;
1743 pool_flags = ceph_pg_pool_flags(osdc->osdmap, ci->i_layout.pool_id);
1744 up_read(&osdc->lock);
1745 if ((map_flags & CEPH_OSDMAP_FULL) ||
1746 (pool_flags & CEPH_POOL_FLAG_FULL)) {
1747 err = -ENOSPC;
1748 goto out;
1749 }
1750
1751 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1752 inode, ceph_vinop(inode), pos, count, i_size_read(inode));
1753 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1754 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1755 else
1756 want = CEPH_CAP_FILE_BUFFER;
1757 got = 0;
1758 err = ceph_get_caps(file, CEPH_CAP_FILE_WR, want, pos + count, &got);
1759 if (err < 0)
1760 goto out;
1761
1762 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1763 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
1764
1765 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1766 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC) ||
1767 (ci->i_ceph_flags & CEPH_I_ERROR_WRITE)) {
1768 struct ceph_snap_context *snapc;
1769 struct iov_iter data;
1770
1771 spin_lock(&ci->i_ceph_lock);
1772 if (__ceph_have_pending_cap_snap(ci)) {
1773 struct ceph_cap_snap *capsnap =
1774 list_last_entry(&ci->i_cap_snaps,
1775 struct ceph_cap_snap,
1776 ci_item);
1777 snapc = ceph_get_snap_context(capsnap->context);
1778 } else {
1779 BUG_ON(!ci->i_head_snapc);
1780 snapc = ceph_get_snap_context(ci->i_head_snapc);
1781 }
1782 spin_unlock(&ci->i_ceph_lock);
1783
1784
1785 data = *from;
1786 if (iocb->ki_flags & IOCB_DIRECT)
1787 written = ceph_direct_read_write(iocb, &data, snapc,
1788 &prealloc_cf);
1789 else
1790 written = ceph_sync_write(iocb, &data, pos, snapc);
1791 if (direct_lock)
1792 ceph_end_io_direct(inode);
1793 else
1794 ceph_end_io_write(inode);
1795 if (written > 0)
1796 iov_iter_advance(from, written);
1797 ceph_put_snap_context(snapc);
1798 } else {
1799
1800
1801
1802
1803
1804
1805
1806 written = generic_perform_write(file, from, pos);
1807 if (likely(written >= 0))
1808 iocb->ki_pos = pos + written;
1809 ceph_end_io_write(inode);
1810 }
1811
1812 if (written >= 0) {
1813 int dirty;
1814
1815 spin_lock(&ci->i_ceph_lock);
1816 ci->i_inline_version = CEPH_INLINE_NONE;
1817 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1818 &prealloc_cf);
1819 spin_unlock(&ci->i_ceph_lock);
1820 if (dirty)
1821 __mark_inode_dirty(inode, dirty);
1822 if (ceph_quota_is_max_bytes_approaching(inode, iocb->ki_pos))
1823 ceph_check_caps(ci, 0, NULL);
1824 }
1825
1826 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1827 inode, ceph_vinop(inode), pos, (unsigned)count,
1828 ceph_cap_string(got));
1829 ceph_put_cap_refs(ci, got);
1830
1831 if (written == -EOLDSNAPC) {
1832 dout("aio_write %p %llx.%llx %llu~%u" "got EOLDSNAPC, retrying\n",
1833 inode, ceph_vinop(inode), pos, (unsigned)count);
1834 goto retry_snap;
1835 }
1836
1837 if (written >= 0) {
1838 if ((map_flags & CEPH_OSDMAP_NEARFULL) ||
1839 (pool_flags & CEPH_POOL_FLAG_NEARFULL))
1840 iocb->ki_flags |= IOCB_DSYNC;
1841 written = generic_write_sync(iocb, written);
1842 }
1843
1844 goto out_unlocked;
1845out:
1846 if (direct_lock)
1847 ceph_end_io_direct(inode);
1848 else
1849 ceph_end_io_write(inode);
1850out_unlocked:
1851 ceph_free_cap_flush(prealloc_cf);
1852 current->backing_dev_info = NULL;
1853 return written ? written : err;
1854}
1855
1856
1857
1858
1859static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
1860{
1861 struct inode *inode = file->f_mapping->host;
1862 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1863 loff_t i_size;
1864 loff_t ret;
1865
1866 inode_lock(inode);
1867
1868 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
1869 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1870 if (ret < 0)
1871 goto out;
1872 }
1873
1874 i_size = i_size_read(inode);
1875 switch (whence) {
1876 case SEEK_END:
1877 offset += i_size;
1878 break;
1879 case SEEK_CUR:
1880
1881
1882
1883
1884
1885
1886 if (offset == 0) {
1887 ret = file->f_pos;
1888 goto out;
1889 }
1890 offset += file->f_pos;
1891 break;
1892 case SEEK_DATA:
1893 if (offset < 0 || offset >= i_size) {
1894 ret = -ENXIO;
1895 goto out;
1896 }
1897 break;
1898 case SEEK_HOLE:
1899 if (offset < 0 || offset >= i_size) {
1900 ret = -ENXIO;
1901 goto out;
1902 }
1903 offset = i_size;
1904 break;
1905 }
1906
1907 ret = vfs_setpos(file, offset, max(i_size, fsc->max_file_size));
1908
1909out:
1910 inode_unlock(inode);
1911 return ret;
1912}
1913
1914static inline void ceph_zero_partial_page(
1915 struct inode *inode, loff_t offset, unsigned size)
1916{
1917 struct page *page;
1918 pgoff_t index = offset >> PAGE_SHIFT;
1919
1920 page = find_lock_page(inode->i_mapping, index);
1921 if (page) {
1922 wait_on_page_writeback(page);
1923 zero_user(page, offset & (PAGE_SIZE - 1), size);
1924 unlock_page(page);
1925 put_page(page);
1926 }
1927}
1928
1929static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1930 loff_t length)
1931{
1932 loff_t nearly = round_up(offset, PAGE_SIZE);
1933 if (offset < nearly) {
1934 loff_t size = nearly - offset;
1935 if (length < size)
1936 size = length;
1937 ceph_zero_partial_page(inode, offset, size);
1938 offset += size;
1939 length -= size;
1940 }
1941 if (length >= PAGE_SIZE) {
1942 loff_t size = round_down(length, PAGE_SIZE);
1943 truncate_pagecache_range(inode, offset, offset + size - 1);
1944 offset += size;
1945 length -= size;
1946 }
1947 if (length)
1948 ceph_zero_partial_page(inode, offset, length);
1949}
1950
1951static int ceph_zero_partial_object(struct inode *inode,
1952 loff_t offset, loff_t *length)
1953{
1954 struct ceph_inode_info *ci = ceph_inode(inode);
1955 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1956 struct ceph_osd_request *req;
1957 int ret = 0;
1958 loff_t zero = 0;
1959 int op;
1960
1961 if (!length) {
1962 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1963 length = &zero;
1964 } else {
1965 op = CEPH_OSD_OP_ZERO;
1966 }
1967
1968 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1969 ceph_vino(inode),
1970 offset, length,
1971 0, 1, op,
1972 CEPH_OSD_FLAG_WRITE,
1973 NULL, 0, 0, false);
1974 if (IS_ERR(req)) {
1975 ret = PTR_ERR(req);
1976 goto out;
1977 }
1978
1979 req->r_mtime = inode->i_mtime;
1980 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1981 if (!ret) {
1982 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1983 if (ret == -ENOENT)
1984 ret = 0;
1985 }
1986 ceph_osdc_put_request(req);
1987
1988out:
1989 return ret;
1990}
1991
1992static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1993{
1994 int ret = 0;
1995 struct ceph_inode_info *ci = ceph_inode(inode);
1996 s32 stripe_unit = ci->i_layout.stripe_unit;
1997 s32 stripe_count = ci->i_layout.stripe_count;
1998 s32 object_size = ci->i_layout.object_size;
1999 u64 object_set_size = object_size * stripe_count;
2000 u64 nearly, t;
2001
2002
2003 nearly = offset + object_set_size - 1;
2004 t = nearly;
2005 nearly -= do_div(t, object_set_size);
2006
2007 while (length && offset < nearly) {
2008 loff_t size = length;
2009 ret = ceph_zero_partial_object(inode, offset, &size);
2010 if (ret < 0)
2011 return ret;
2012 offset += size;
2013 length -= size;
2014 }
2015 while (length >= object_set_size) {
2016 int i;
2017 loff_t pos = offset;
2018 for (i = 0; i < stripe_count; ++i) {
2019 ret = ceph_zero_partial_object(inode, pos, NULL);
2020 if (ret < 0)
2021 return ret;
2022 pos += stripe_unit;
2023 }
2024 offset += object_set_size;
2025 length -= object_set_size;
2026 }
2027 while (length) {
2028 loff_t size = length;
2029 ret = ceph_zero_partial_object(inode, offset, &size);
2030 if (ret < 0)
2031 return ret;
2032 offset += size;
2033 length -= size;
2034 }
2035 return ret;
2036}
2037
2038static long ceph_fallocate(struct file *file, int mode,
2039 loff_t offset, loff_t length)
2040{
2041 struct ceph_file_info *fi = file->private_data;
2042 struct inode *inode = file_inode(file);
2043 struct ceph_inode_info *ci = ceph_inode(inode);
2044 struct ceph_cap_flush *prealloc_cf;
2045 int want, got = 0;
2046 int dirty;
2047 int ret = 0;
2048 loff_t endoff = 0;
2049 loff_t size;
2050
2051 if (mode != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2052 return -EOPNOTSUPP;
2053
2054 if (!S_ISREG(inode->i_mode))
2055 return -EOPNOTSUPP;
2056
2057 prealloc_cf = ceph_alloc_cap_flush();
2058 if (!prealloc_cf)
2059 return -ENOMEM;
2060
2061 inode_lock(inode);
2062
2063 if (ceph_snap(inode) != CEPH_NOSNAP) {
2064 ret = -EROFS;
2065 goto unlock;
2066 }
2067
2068 if (ci->i_inline_version != CEPH_INLINE_NONE) {
2069 ret = ceph_uninline_data(file, NULL);
2070 if (ret < 0)
2071 goto unlock;
2072 }
2073
2074 size = i_size_read(inode);
2075
2076
2077 if (offset >= size)
2078 goto unlock;
2079 if ((offset + length) > size)
2080 length = size - offset;
2081
2082 if (fi->fmode & CEPH_FILE_MODE_LAZY)
2083 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
2084 else
2085 want = CEPH_CAP_FILE_BUFFER;
2086
2087 ret = ceph_get_caps(file, CEPH_CAP_FILE_WR, want, endoff, &got);
2088 if (ret < 0)
2089 goto unlock;
2090
2091 ceph_zero_pagecache_range(inode, offset, length);
2092 ret = ceph_zero_objects(inode, offset, length);
2093
2094 if (!ret) {
2095 spin_lock(&ci->i_ceph_lock);
2096 ci->i_inline_version = CEPH_INLINE_NONE;
2097 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
2098 &prealloc_cf);
2099 spin_unlock(&ci->i_ceph_lock);
2100 if (dirty)
2101 __mark_inode_dirty(inode, dirty);
2102 }
2103
2104 ceph_put_cap_refs(ci, got);
2105unlock:
2106 inode_unlock(inode);
2107 ceph_free_cap_flush(prealloc_cf);
2108 return ret;
2109}
2110
2111
2112
2113
2114
2115
2116static int get_rd_wr_caps(struct file *src_filp, int *src_got,
2117 struct file *dst_filp,
2118 loff_t dst_endoff, int *dst_got)
2119{
2120 int ret = 0;
2121 bool retrying = false;
2122
2123retry_caps:
2124 ret = ceph_get_caps(dst_filp, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER,
2125 dst_endoff, dst_got);
2126 if (ret < 0)
2127 return ret;
2128
2129
2130
2131
2132
2133
2134 ret = ceph_try_get_caps(file_inode(src_filp),
2135 CEPH_CAP_FILE_RD, CEPH_CAP_FILE_SHARED,
2136 false, src_got);
2137 if (ret <= 0) {
2138
2139 ceph_put_cap_refs(ceph_inode(file_inode(dst_filp)), *dst_got);
2140 if (retrying) {
2141 if (!ret)
2142
2143 ret = -EAGAIN;
2144 return ret;
2145 }
2146 ret = ceph_get_caps(src_filp, CEPH_CAP_FILE_RD,
2147 CEPH_CAP_FILE_SHARED, -1, src_got);
2148 if (ret < 0)
2149 return ret;
2150
2151 ceph_put_cap_refs(ceph_inode(file_inode(src_filp)), *src_got);
2152 retrying = true;
2153 goto retry_caps;
2154 }
2155 return ret;
2156}
2157
2158static void put_rd_wr_caps(struct ceph_inode_info *src_ci, int src_got,
2159 struct ceph_inode_info *dst_ci, int dst_got)
2160{
2161 ceph_put_cap_refs(src_ci, src_got);
2162 ceph_put_cap_refs(dst_ci, dst_got);
2163}
2164
2165
2166
2167
2168
2169
2170
2171static int is_file_size_ok(struct inode *src_inode, struct inode *dst_inode,
2172 loff_t src_off, loff_t dst_off, size_t len)
2173{
2174 loff_t size, endoff;
2175
2176 size = i_size_read(src_inode);
2177
2178
2179
2180
2181
2182
2183 if (src_off + len > size) {
2184 dout("Copy beyond EOF (%llu + %zu > %llu)\n",
2185 src_off, len, size);
2186 return -EOPNOTSUPP;
2187 }
2188 size = i_size_read(dst_inode);
2189
2190 endoff = dst_off + len;
2191 if (inode_newsize_ok(dst_inode, endoff))
2192 return -EOPNOTSUPP;
2193
2194 if (ceph_quota_is_max_bytes_exceeded(dst_inode, endoff))
2195 return -EDQUOT;
2196
2197 return 0;
2198}
2199
2200static ssize_t ceph_do_objects_copy(struct ceph_inode_info *src_ci, u64 *src_off,
2201 struct ceph_inode_info *dst_ci, u64 *dst_off,
2202 struct ceph_fs_client *fsc,
2203 size_t len, unsigned int flags)
2204{
2205 struct ceph_object_locator src_oloc, dst_oloc;
2206 struct ceph_object_id src_oid, dst_oid;
2207 size_t bytes = 0;
2208 u64 src_objnum, src_objoff, dst_objnum, dst_objoff;
2209 u32 src_objlen, dst_objlen;
2210 u32 object_size = src_ci->i_layout.object_size;
2211 int ret;
2212
2213 src_oloc.pool = src_ci->i_layout.pool_id;
2214 src_oloc.pool_ns = ceph_try_get_string(src_ci->i_layout.pool_ns);
2215 dst_oloc.pool = dst_ci->i_layout.pool_id;
2216 dst_oloc.pool_ns = ceph_try_get_string(dst_ci->i_layout.pool_ns);
2217
2218 while (len >= object_size) {
2219 ceph_calc_file_object_mapping(&src_ci->i_layout, *src_off,
2220 object_size, &src_objnum,
2221 &src_objoff, &src_objlen);
2222 ceph_calc_file_object_mapping(&dst_ci->i_layout, *dst_off,
2223 object_size, &dst_objnum,
2224 &dst_objoff, &dst_objlen);
2225 ceph_oid_init(&src_oid);
2226 ceph_oid_printf(&src_oid, "%llx.%08llx",
2227 src_ci->i_vino.ino, src_objnum);
2228 ceph_oid_init(&dst_oid);
2229 ceph_oid_printf(&dst_oid, "%llx.%08llx",
2230 dst_ci->i_vino.ino, dst_objnum);
2231
2232 ret = ceph_osdc_copy_from(&fsc->client->osdc,
2233 src_ci->i_vino.snap, 0,
2234 &src_oid, &src_oloc,
2235 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2236 CEPH_OSD_OP_FLAG_FADVISE_NOCACHE,
2237 &dst_oid, &dst_oloc,
2238 CEPH_OSD_OP_FLAG_FADVISE_SEQUENTIAL |
2239 CEPH_OSD_OP_FLAG_FADVISE_DONTNEED,
2240 dst_ci->i_truncate_seq,
2241 dst_ci->i_truncate_size,
2242 CEPH_OSD_COPY_FROM_FLAG_TRUNCATE_SEQ);
2243 if (ret) {
2244 if (ret == -EOPNOTSUPP) {
2245 fsc->have_copy_from2 = false;
2246 pr_notice("OSDs don't support copy-from2; disabling copy offload\n");
2247 }
2248 dout("ceph_osdc_copy_from returned %d\n", ret);
2249 if (!bytes)
2250 bytes = ret;
2251 goto out;
2252 }
2253 len -= object_size;
2254 bytes += object_size;
2255 *src_off += object_size;
2256 *dst_off += object_size;
2257 }
2258
2259out:
2260 ceph_oloc_destroy(&src_oloc);
2261 ceph_oloc_destroy(&dst_oloc);
2262 return bytes;
2263}
2264
2265static ssize_t __ceph_copy_file_range(struct file *src_file, loff_t src_off,
2266 struct file *dst_file, loff_t dst_off,
2267 size_t len, unsigned int flags)
2268{
2269 struct inode *src_inode = file_inode(src_file);
2270 struct inode *dst_inode = file_inode(dst_file);
2271 struct ceph_inode_info *src_ci = ceph_inode(src_inode);
2272 struct ceph_inode_info *dst_ci = ceph_inode(dst_inode);
2273 struct ceph_cap_flush *prealloc_cf;
2274 struct ceph_fs_client *src_fsc = ceph_inode_to_client(src_inode);
2275 loff_t size;
2276 ssize_t ret = -EIO, bytes;
2277 u64 src_objnum, dst_objnum, src_objoff, dst_objoff;
2278 u32 src_objlen, dst_objlen;
2279 int src_got = 0, dst_got = 0, err, dirty;
2280
2281 if (src_inode->i_sb != dst_inode->i_sb) {
2282 struct ceph_fs_client *dst_fsc = ceph_inode_to_client(dst_inode);
2283
2284 if (ceph_fsid_compare(&src_fsc->client->fsid,
2285 &dst_fsc->client->fsid)) {
2286 dout("Copying files across clusters: src: %pU dst: %pU\n",
2287 &src_fsc->client->fsid, &dst_fsc->client->fsid);
2288 return -EXDEV;
2289 }
2290 }
2291 if (ceph_snap(dst_inode) != CEPH_NOSNAP)
2292 return -EROFS;
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302 if (ceph_test_mount_opt(src_fsc, NOCOPYFROM))
2303 return -EOPNOTSUPP;
2304
2305 if (!src_fsc->have_copy_from2)
2306 return -EOPNOTSUPP;
2307
2308
2309
2310
2311
2312
2313 if ((src_ci->i_layout.stripe_unit != dst_ci->i_layout.stripe_unit) ||
2314 (src_ci->i_layout.stripe_count != 1) ||
2315 (dst_ci->i_layout.stripe_count != 1) ||
2316 (src_ci->i_layout.object_size != dst_ci->i_layout.object_size)) {
2317 dout("Invalid src/dst files layout\n");
2318 return -EOPNOTSUPP;
2319 }
2320
2321 if (len < src_ci->i_layout.object_size)
2322 return -EOPNOTSUPP;
2323
2324 prealloc_cf = ceph_alloc_cap_flush();
2325 if (!prealloc_cf)
2326 return -ENOMEM;
2327
2328
2329 ret = file_write_and_wait_range(src_file, src_off, (src_off + len));
2330 if (ret < 0) {
2331 dout("failed to write src file (%zd)\n", ret);
2332 goto out;
2333 }
2334 ret = file_write_and_wait_range(dst_file, dst_off, (dst_off + len));
2335 if (ret < 0) {
2336 dout("failed to write dst file (%zd)\n", ret);
2337 goto out;
2338 }
2339
2340
2341
2342
2343
2344
2345 err = get_rd_wr_caps(src_file, &src_got,
2346 dst_file, (dst_off + len), &dst_got);
2347 if (err < 0) {
2348 dout("get_rd_wr_caps returned %d\n", err);
2349 ret = -EOPNOTSUPP;
2350 goto out;
2351 }
2352
2353 ret = is_file_size_ok(src_inode, dst_inode, src_off, dst_off, len);
2354 if (ret < 0)
2355 goto out_caps;
2356
2357
2358 ret = invalidate_inode_pages2_range(dst_inode->i_mapping,
2359 dst_off >> PAGE_SHIFT,
2360 (dst_off + len) >> PAGE_SHIFT);
2361 if (ret < 0) {
2362 dout("Failed to invalidate inode pages (%zd)\n", ret);
2363 ret = 0;
2364 }
2365 ceph_calc_file_object_mapping(&src_ci->i_layout, src_off,
2366 src_ci->i_layout.object_size,
2367 &src_objnum, &src_objoff, &src_objlen);
2368 ceph_calc_file_object_mapping(&dst_ci->i_layout, dst_off,
2369 dst_ci->i_layout.object_size,
2370 &dst_objnum, &dst_objoff, &dst_objlen);
2371
2372 if (src_objoff != dst_objoff) {
2373 ret = -EOPNOTSUPP;
2374 goto out_caps;
2375 }
2376
2377
2378
2379
2380
2381
2382 if (src_objoff) {
2383 dout("Initial partial copy of %u bytes\n", src_objlen);
2384
2385
2386
2387
2388
2389 put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
2390 ret = do_splice_direct(src_file, &src_off, dst_file,
2391 &dst_off, src_objlen, flags);
2392
2393 if (ret < src_objlen) {
2394 dout("Failed partial copy (%zd)\n", ret);
2395 goto out;
2396 }
2397 len -= ret;
2398 err = get_rd_wr_caps(src_file, &src_got,
2399 dst_file, (dst_off + len), &dst_got);
2400 if (err < 0)
2401 goto out;
2402 err = is_file_size_ok(src_inode, dst_inode,
2403 src_off, dst_off, len);
2404 if (err < 0)
2405 goto out_caps;
2406 }
2407
2408 size = i_size_read(dst_inode);
2409 bytes = ceph_do_objects_copy(src_ci, &src_off, dst_ci, &dst_off,
2410 src_fsc, len, flags);
2411 if (bytes <= 0) {
2412 if (!ret)
2413 ret = bytes;
2414 goto out_caps;
2415 }
2416 dout("Copied %zu bytes out of %zu\n", bytes, len);
2417 len -= bytes;
2418 ret += bytes;
2419
2420 file_update_time(dst_file);
2421 inode_inc_iversion_raw(dst_inode);
2422
2423 if (dst_off > size) {
2424
2425 if (ceph_inode_set_size(dst_inode, dst_off) ||
2426 ceph_quota_is_max_bytes_approaching(dst_inode, dst_off))
2427 ceph_check_caps(dst_ci, CHECK_CAPS_AUTHONLY, NULL);
2428 }
2429
2430 spin_lock(&dst_ci->i_ceph_lock);
2431 dst_ci->i_inline_version = CEPH_INLINE_NONE;
2432 dirty = __ceph_mark_dirty_caps(dst_ci, CEPH_CAP_FILE_WR, &prealloc_cf);
2433 spin_unlock(&dst_ci->i_ceph_lock);
2434 if (dirty)
2435 __mark_inode_dirty(dst_inode, dirty);
2436
2437out_caps:
2438 put_rd_wr_caps(src_ci, src_got, dst_ci, dst_got);
2439
2440
2441
2442
2443
2444 if (len && (len < src_ci->i_layout.object_size)) {
2445 dout("Final partial copy of %zu bytes\n", len);
2446 bytes = do_splice_direct(src_file, &src_off, dst_file,
2447 &dst_off, len, flags);
2448 if (bytes > 0)
2449 ret += bytes;
2450 else
2451 dout("Failed partial copy (%zd)\n", bytes);
2452 }
2453
2454out:
2455 ceph_free_cap_flush(prealloc_cf);
2456
2457 return ret;
2458}
2459
2460static ssize_t ceph_copy_file_range(struct file *src_file, loff_t src_off,
2461 struct file *dst_file, loff_t dst_off,
2462 size_t len, unsigned int flags)
2463{
2464 ssize_t ret;
2465
2466 ret = __ceph_copy_file_range(src_file, src_off, dst_file, dst_off,
2467 len, flags);
2468
2469 if (ret == -EOPNOTSUPP || ret == -EXDEV)
2470 ret = generic_copy_file_range(src_file, src_off, dst_file,
2471 dst_off, len, flags);
2472 return ret;
2473}
2474
2475const struct file_operations ceph_file_fops = {
2476 .open = ceph_open,
2477 .release = ceph_release,
2478 .llseek = ceph_llseek,
2479 .read_iter = ceph_read_iter,
2480 .write_iter = ceph_write_iter,
2481 .mmap = ceph_mmap,
2482 .fsync = ceph_fsync,
2483 .lock = ceph_lock,
2484 .setlease = simple_nosetlease,
2485 .flock = ceph_flock,
2486 .splice_read = generic_file_splice_read,
2487 .splice_write = iter_file_splice_write,
2488 .unlocked_ioctl = ceph_ioctl,
2489 .compat_ioctl = compat_ptr_ioctl,
2490 .fallocate = ceph_fallocate,
2491 .copy_file_range = ceph_copy_file_range,
2492};
2493
