]>
git.proxmox.com Git - ceph.git/blob - ceph/src/client/Inode.cc
1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
2 // vim: ts=8 sw=2 smarttab
9 #include "MetaSession.h"
10 #include "ClientSnapRealm.h"
11 #include "Delegation.h"
13 #include "mds/flock.h"
17 delay_cap_item
.remove_myself();
18 dirty_cap_item
.remove_myself();
19 snaprealm_item
.remove_myself();
22 snapdir_parent
->flags
&= ~I_SNAPDIR_OPEN
;
23 snapdir_parent
.reset();
26 if (!oset
.objects
.empty()) {
27 lsubdout(client
->cct
, client
, 0) << __func__
<< ": leftover objects on inode 0x"
28 << std::hex
<< ino
<< std::dec
<< dendl
;
29 ceph_assert(oset
.objects
.empty());
32 if (!delegations
.empty()) {
33 lsubdout(client
->cct
, client
, 0) << __func__
<< ": leftover delegations on inode 0x"
34 << std::hex
<< ino
<< std::dec
<< dendl
;
35 ceph_assert(delegations
.empty());
39 ostream
& operator<<(ostream
&out
, const Inode
&in
)
41 out
<< in
.vino() << "("
42 << "faked_ino=" << in
.faked_ino
44 << " ll_ref=" << in
.ll_ref
45 << " cap_refs=" << in
.cap_refs
46 << " open=" << in
.open_by_mode
47 << " mode=" << oct
<< in
.mode
<< dec
48 << " size=" << in
.size
<< "/" << in
.max_size
49 << " nlink=" << in
.nlink
50 << " btime=" << in
.btime
51 << " mtime=" << in
.mtime
52 << " ctime=" << in
.ctime
53 << " caps=" << ccap_string(in
.caps_issued());
54 if (!in
.caps
.empty()) {
57 for (const auto &pair
: in
.caps
) {
60 out
<< pair
.first
<< '=' << ccap_string(pair
.second
.issued
);
66 out
<< " dirty_caps=" << ccap_string(in
.dirty_caps
);
68 out
<< " flushing_caps=" << ccap_string(in
.flushing_caps
);
70 if (in
.flags
& I_COMPLETE
)
74 out
<< " " << in
.oset
;
76 if (!in
.dentries
.empty())
77 out
<< " parents=" << in
.dentries
;
79 if (in
.is_dir() && in
.has_dir_layout())
80 out
<< " has_dir_layout";
82 if (in
.quota
.is_enable())
83 out
<< " " << in
.quota
;
85 out
<< ' ' << &in
<< ")";
90 void Inode::make_long_path(filepath
& p
)
92 if (!dentries
.empty()) {
93 Dentry
*dn
= get_first_parent();
94 ceph_assert(dn
->dir
&& dn
->dir
->parent_inode
);
95 dn
->dir
->parent_inode
->make_long_path(p
);
96 p
.push_dentry(dn
->name
);
97 } else if (snapdir_parent
) {
98 make_nosnap_relative_path(p
);
103 void Inode::make_short_path(filepath
& p
)
105 if (!dentries
.empty()) {
106 Dentry
*dn
= get_first_parent();
107 ceph_assert(dn
->dir
&& dn
->dir
->parent_inode
);
108 p
= filepath(dn
->name
, dn
->dir
->parent_inode
->ino
);
109 } else if (snapdir_parent
) {
110 make_nosnap_relative_path(p
);
116 * make a filepath suitable for an mds request:
117 * - if we are non-snapped/live, the ino is sufficient, e.g. #1234
118 * - if we are snapped, make filepath relative to first non-snapped parent.
120 void Inode::make_nosnap_relative_path(filepath
& p
)
122 if (snapid
== CEPH_NOSNAP
) {
124 } else if (snapdir_parent
) {
125 snapdir_parent
->make_nosnap_relative_path(p
);
127 p
.push_dentry(empty
);
128 } else if (!dentries
.empty()) {
129 Dentry
*dn
= get_first_parent();
130 ceph_assert(dn
->dir
&& dn
->dir
->parent_inode
);
131 dn
->dir
->parent_inode
->make_nosnap_relative_path(p
);
132 p
.push_dentry(dn
->name
);
138 void Inode::get_open_ref(int mode
)
140 open_by_mode
[mode
]++;
141 break_deleg(!(mode
& CEPH_FILE_MODE_WR
));
144 bool Inode::put_open_ref(int mode
)
146 //cout << "open_by_mode[" << mode << "] " << open_by_mode[mode] << " -> " << (open_by_mode[mode]-1) << std::endl;
147 auto& ref
= open_by_mode
.at(mode
);
148 ceph_assert(ref
> 0);
154 void Inode::get_cap_ref(int cap
)
161 //cout << "inode " << *this << " get " << cap_string(c) << " " << (cap_refs[c]-1) << " -> " << cap_refs[c] << std::endl;
168 int Inode::put_cap_ref(int cap
)
175 if (cap_refs
[c
] <= 0) {
176 lderr(client
->cct
) << "put_cap_ref " << ccap_string(c
) << " went negative on " << *this << dendl
;
177 ceph_assert(cap_refs
[c
] > 0);
179 if (--cap_refs
[c
] == 0)
181 //cout << "inode " << *this << " put " << cap_string(c) << " " << (cap_refs[c]+1) << " -> " << cap_refs[c] << std::endl;
189 bool Inode::is_any_caps()
191 return !caps
.empty() || snap_caps
;
194 bool Inode::cap_is_valid(const Cap
&cap
) const
196 /*cout << "cap_gen " << cap->session-> cap_gen << std::endl
197 << "session gen " << cap->gen << std::endl
198 << "cap expire " << cap->session->cap_ttl << std::endl
199 << "cur time " << ceph_clock_now(cct) << std::endl;*/
200 if ((cap
.session
->cap_gen
<= cap
.gen
)
201 && (ceph_clock_now() < cap
.session
->cap_ttl
)) {
207 int Inode::caps_issued(int *implemented
) const
211 for (const auto &pair
: caps
) {
212 const Cap
&cap
= pair
.second
;
213 if (cap_is_valid(cap
)) {
215 i
|= cap
.implemented
;
218 // exclude caps issued by non-auth MDS, but are been revoking by
219 // the auth MDS. The non-auth MDS should be revoking/exporting
220 // these caps, but the message is delayed.
222 c
&= ~auth_cap
->implemented
| auth_cap
->issued
;
229 void Inode::try_touch_cap(mds_rank_t mds
)
231 auto it
= caps
.find(mds
);
232 if (it
!= caps
.end()) {
238 * caps_issued_mask - check whether we have all of the caps in the mask
239 * @mask: mask to check against
240 * @allow_impl: whether the caller can also use caps that are implemented but not issued
242 * This is the bog standard "check whether we have the required caps" operation.
243 * Typically, we only check against the capset that is currently "issued".
244 * In other words, we ignore caps that have been revoked but not yet released.
246 * Some callers (particularly those doing attribute retrieval) can also make
247 * use of the full set of "implemented" caps to satisfy requests from the
250 * Those callers should refrain from taking new references to implemented
253 bool Inode::caps_issued_mask(unsigned mask
, bool allow_impl
)
258 if ((c
& mask
) == mask
)
262 cap_is_valid(*auth_cap
) &&
263 (auth_cap
->issued
& mask
) == mask
) {
268 for (auto &pair
: caps
) {
269 Cap
&cap
= pair
.second
;
270 if (cap_is_valid(cap
)) {
271 if ((cap
.issued
& mask
) == mask
) {
276 i
|= cap
.implemented
;
283 if ((c
& mask
) == mask
) {
284 // bah.. touch them all
285 for (auto &pair
: caps
) {
293 int Inode::caps_used()
296 for (map
<int,int>::iterator p
= cap_refs
.begin();
304 int Inode::caps_file_wanted()
307 for (map
<int,int>::iterator p
= open_by_mode
.begin();
308 p
!= open_by_mode
.end();
311 want
|= ceph_caps_for_mode(p
->first
);
315 int Inode::caps_wanted()
317 int want
= caps_file_wanted() | caps_used();
318 if (want
& CEPH_CAP_FILE_BUFFER
)
319 want
|= CEPH_CAP_FILE_EXCL
;
323 int Inode::caps_mds_wanted()
326 for (const auto &pair
: caps
) {
327 want
|= pair
.second
.wanted
;
332 int Inode::caps_dirty()
334 return dirty_caps
| flushing_caps
;
337 const UserPerm
* Inode::get_best_perms()
339 const UserPerm
*perms
= NULL
;
340 for (const auto &pair
: caps
) {
341 const UserPerm
& iperm
= pair
.second
.latest_perms
;
342 if (!perms
) { // we don't have any, take what's present
344 } else if (iperm
.uid() == uid
) {
345 if (iperm
.gid() == gid
) { // we have the best possible, return
348 if (perms
->uid() != uid
) { // take uid > gid every time
351 } else if (perms
->uid() != uid
&& iperm
.gid() == gid
) {
352 perms
= &iperm
; // a matching gid is better than nothing
358 bool Inode::have_valid_size()
360 // RD+RDCACHE or WR+WRBUFFER => valid size
361 if (caps_issued() & (CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_EXCL
))
366 // open Dir for an inode. if it's not open, allocated it (and pin dentry in memory).
367 Dir
*Inode::open_dir()
371 lsubdout(client
->cct
, client
, 15) << "open_dir " << dir
<< " on " << this << dendl
;
372 ceph_assert(dentries
.size() < 2); // dirs can't be hard-linked
373 if (!dentries
.empty())
374 get_first_parent()->get(); // pin dentry
380 bool Inode::check_mode(const UserPerm
& perms
, unsigned want
)
382 if (uid
== perms
.uid()) {
383 // if uid is owner, owner entry determines access
385 } else if (perms
.gid_in_groups(gid
)) {
386 // if a gid or sgid matches the owning group, group entry determines access
390 return (mode
& want
) == want
;
395 lsubdout(client
->cct
, client
, 15) << "inode.get on " << this << " " << ino
<< '.' << snapid
396 << " now " << _ref
<< dendl
;
399 //private method to put a reference; see Client::put_inode()
400 int Inode::_put(int n
) {
402 lsubdout(client
->cct
, client
, 15) << "inode.put on " << this << " " << ino
<< '.' << snapid
403 << " now " << _ref
<< dendl
;
404 ceph_assert(_ref
>= 0);
409 void Inode::dump(Formatter
*f
) const
411 f
->dump_stream("ino") << ino
;
412 f
->dump_stream("snapid") << snapid
;
414 f
->dump_unsigned("rdev", rdev
);
415 f
->dump_stream("ctime") << ctime
;
416 f
->dump_stream("btime") << btime
;
417 f
->dump_stream("mode") << '0' << std::oct
<< mode
<< std::dec
;
418 f
->dump_unsigned("uid", uid
);
419 f
->dump_unsigned("gid", gid
);
420 f
->dump_int("nlink", nlink
);
422 f
->dump_unsigned("size", size
);
423 f
->dump_unsigned("max_size", max_size
);
424 f
->dump_unsigned("truncate_seq", truncate_seq
);
425 f
->dump_unsigned("truncate_size", truncate_size
);
426 f
->dump_stream("mtime") << mtime
;
427 f
->dump_stream("atime") << atime
;
428 f
->dump_unsigned("time_warp_seq", time_warp_seq
);
429 f
->dump_unsigned("change_attr", change_attr
);
431 f
->dump_object("layout", layout
);
433 f
->open_object_section("dir_layout");
434 ::dump(dir_layout
, f
);
437 f
->dump_bool("complete", flags
& I_COMPLETE
);
438 f
->dump_bool("ordered", flags
& I_DIR_ORDERED
);
440 /* FIXME when wip-mds-encoding is merged ***
441 f->open_object_section("dir_stat");
445 f->open_object_section("rstat");
451 f
->dump_unsigned("version", version
);
452 f
->dump_unsigned("xattr_version", xattr_version
);
453 f
->dump_unsigned("flags", flags
);
456 f
->dump_int("dir_hashed", (int)dir_hashed
);
457 f
->dump_int("dir_replicated", (int)dir_replicated
);
460 f
->open_array_section("caps");
461 for (const auto &pair
: caps
) {
462 f
->open_object_section("cap");
463 f
->dump_int("mds", pair
.first
);
464 if (&pair
.second
== auth_cap
)
465 f
->dump_int("auth", 1);
471 f
->dump_int("auth_cap", auth_cap
->session
->mds_num
);
473 f
->dump_stream("dirty_caps") << ccap_string(dirty_caps
);
475 f
->dump_stream("flushings_caps") << ccap_string(flushing_caps
);
476 f
->open_object_section("flushing_cap_tid");
477 for (map
<ceph_tid_t
, int>::const_iterator p
= flushing_cap_tids
.begin();
478 p
!= flushing_cap_tids
.end();
480 string
n(ccap_string(p
->second
));
481 f
->dump_unsigned(n
.c_str(), p
->first
);
485 f
->dump_int("shared_gen", shared_gen
);
486 f
->dump_int("cache_gen", cache_gen
);
488 f
->dump_int("snap_caps", snap_caps
);
489 f
->dump_int("snap_cap_refs", snap_cap_refs
);
492 f
->dump_stream("hold_caps_until") << hold_caps_until
;
495 f
->open_object_section("snaprealm");
499 if (!cap_snaps
.empty()) {
500 for (const auto &p
: cap_snaps
) {
501 f
->open_object_section("cap_snap");
502 f
->dump_stream("follows") << p
.first
;
509 if (!open_by_mode
.empty()) {
510 f
->open_array_section("open_by_mode");
511 for (map
<int,int>::const_iterator p
= open_by_mode
.begin(); p
!= open_by_mode
.end(); ++p
) {
512 f
->open_object_section("ref");
513 f
->dump_int("mode", p
->first
);
514 f
->dump_int("refs", p
->second
);
519 if (!cap_refs
.empty()) {
520 f
->open_array_section("cap_refs");
521 for (map
<int,int>::const_iterator p
= cap_refs
.begin(); p
!= cap_refs
.end(); ++p
) {
522 f
->open_object_section("cap_ref");
523 f
->dump_stream("cap") << ccap_string(p
->first
);
524 f
->dump_int("refs", p
->second
);
530 f
->dump_unsigned("reported_size", reported_size
);
531 if (wanted_max_size
!= max_size
)
532 f
->dump_unsigned("wanted_max_size", wanted_max_size
);
533 if (requested_max_size
!= max_size
)
534 f
->dump_unsigned("requested_max_size", requested_max_size
);
536 f
->dump_int("ref", _ref
);
537 f
->dump_int("ll_ref", ll_ref
);
539 if (!dentries
.empty()) {
540 f
->open_array_section("parents");
541 for (const auto &&dn
: dentries
) {
542 f
->open_object_section("dentry");
543 f
->dump_stream("dir_ino") << dn
->dir
->parent_inode
->ino
;
544 f
->dump_string("name", dn
->name
);
551 void Cap::dump(Formatter
*f
) const
553 f
->dump_int("mds", session
->mds_num
);
554 f
->dump_stream("ino") << inode
.ino
;
555 f
->dump_unsigned("cap_id", cap_id
);
556 f
->dump_stream("issued") << ccap_string(issued
);
557 if (implemented
!= issued
)
558 f
->dump_stream("implemented") << ccap_string(implemented
);
559 f
->dump_stream("wanted") << ccap_string(wanted
);
560 f
->dump_unsigned("seq", seq
);
561 f
->dump_unsigned("issue_seq", issue_seq
);
562 f
->dump_unsigned("mseq", mseq
);
563 f
->dump_unsigned("gen", gen
);
566 void CapSnap::dump(Formatter
*f
) const
568 f
->dump_stream("ino") << in
->ino
;
569 f
->dump_stream("issued") << ccap_string(issued
);
570 f
->dump_stream("dirty") << ccap_string(dirty
);
571 f
->dump_unsigned("size", size
);
572 f
->dump_stream("ctime") << ctime
;
573 f
->dump_stream("mtime") << mtime
;
574 f
->dump_stream("atime") << atime
;
575 f
->dump_int("time_warp_seq", time_warp_seq
);
576 f
->dump_stream("mode") << '0' << std::oct
<< mode
<< std::dec
;
577 f
->dump_unsigned("uid", uid
);
578 f
->dump_unsigned("gid", gid
);
579 if (!xattrs
.empty()) {
580 f
->open_object_section("xattr_lens");
581 for (map
<string
,bufferptr
>::const_iterator p
= xattrs
.begin(); p
!= xattrs
.end(); ++p
)
582 f
->dump_int(p
->first
.c_str(), p
->second
.length());
585 f
->dump_unsigned("xattr_version", xattr_version
);
586 f
->dump_int("writing", (int)writing
);
587 f
->dump_int("dirty_data", (int)dirty_data
);
588 f
->dump_unsigned("flush_tid", flush_tid
);
591 void Inode::set_async_err(int r
)
593 for (const auto &fh
: fhs
) {
598 bool Inode::has_recalled_deleg()
600 if (delegations
.empty())
603 // Either all delegations are recalled or none are. Just check the first.
604 Delegation
& deleg
= delegations
.front();
605 return deleg
.is_recalled();
608 void Inode::recall_deleg(bool skip_read
)
610 if (delegations
.empty())
614 for (list
<Delegation
>::iterator d
= delegations
.begin();
615 d
!= delegations
.end(); ++d
) {
617 Delegation
& deleg
= *d
;
618 deleg
.recall(skip_read
);
622 bool Inode::delegations_broken(bool skip_read
)
624 if (delegations
.empty()) {
625 lsubdout(client
->cct
, client
, 10) <<
626 __func__
<< ": delegations empty on " << *this << dendl
;
631 Delegation
& deleg
= delegations
.front();
632 lsubdout(client
->cct
, client
, 10) <<
633 __func__
<< ": read delegs only on " << *this << dendl
;
634 if (deleg
.get_type() == CEPH_FILE_MODE_RD
) {
638 lsubdout(client
->cct
, client
, 10) <<
639 __func__
<< ": not broken" << *this << dendl
;
643 void Inode::break_deleg(bool skip_read
)
645 lsubdout(client
->cct
, client
, 10) <<
646 __func__
<< ": breaking delegs on " << *this << dendl
;
648 recall_deleg(skip_read
);
650 while (!delegations_broken(skip_read
))
651 client
->wait_on_list(waitfor_deleg
);
655 * set_deleg: request a delegation on an open Fh
656 * @fh: filehandle on which to acquire it
657 * @type: delegation request type
658 * @cb: delegation recall callback function
659 * @priv: private pointer to be passed to callback
661 * Attempt to acquire a delegation on an open file handle. If there are no
662 * conflicts and we have the right caps, allocate a new delegation, fill it
663 * out and return 0. Return an error if we can't get one for any reason.
665 int Inode::set_deleg(Fh
*fh
, unsigned type
, ceph_deleg_cb_t cb
, void *priv
)
667 lsubdout(client
->cct
, client
, 10) <<
668 __func__
<< ": inode " << *this << dendl
;
671 * 0 deleg timeout means that they haven't been explicitly enabled. Don't
672 * allow it, with an unusual error to make it clear.
674 if (!client
->get_deleg_timeout())
677 // Just say no if we have any recalled delegs still outstanding
678 if (has_recalled_deleg()) {
679 lsubdout(client
->cct
, client
, 10) << __func__
<<
680 ": has_recalled_deleg" << dendl
;
684 // check vs. currently open files on this inode
686 case CEPH_DELEGATION_RD
:
687 if (open_count_for_write()) {
688 lsubdout(client
->cct
, client
, 10) << __func__
<<
689 ": open for write" << dendl
;
693 case CEPH_DELEGATION_WR
:
694 if (open_count() > 1) {
695 lsubdout(client
->cct
, client
, 10) << __func__
<< ": open" << dendl
;
704 * A delegation is essentially a long-held container for cap references that
705 * we delegate to the client until recalled. The caps required depend on the
706 * type of delegation (read vs. rw). This is entirely an opportunistic thing.
707 * If we don't have the necessary caps for the delegation, then we just don't
710 * In principle we could request the caps from the MDS, but a delegation is
711 * usually requested just after an open. If we don't have the necessary caps
712 * already, then it's likely that there is some sort of conflicting access.
714 * In the future, we may need to add a way to have this request caps more
715 * aggressively -- for instance, to handle WANT_DELEGATION for NFSv4.1+.
717 int need
= ceph_deleg_caps_for_type(type
);
718 if (!caps_issued_mask(need
)) {
719 lsubdout(client
->cct
, client
, 10) << __func__
<< ": cap mismatch, have="
720 << ccap_string(caps_issued()) << " need=" << ccap_string(need
) << dendl
;
724 for (list
<Delegation
>::iterator d
= delegations
.begin();
725 d
!= delegations
.end(); ++d
) {
726 Delegation
& deleg
= *d
;
727 if (deleg
.get_fh() == fh
) {
728 deleg
.reinit(type
, cb
, priv
);
733 delegations
.emplace_back(fh
, type
, cb
, priv
);
738 * unset_deleg - remove a delegation that was previously set
739 * @fh: file handle to clear delegation of
741 * Unlink delegation from the Inode (if there is one), put caps and free it.
743 void Inode::unset_deleg(Fh
*fh
)
745 for (list
<Delegation
>::iterator d
= delegations
.begin();
746 d
!= delegations
.end(); ++d
) {
747 Delegation
& deleg
= *d
;
748 if (deleg
.get_fh() == fh
) {
749 delegations
.erase(d
);
750 client
->signal_cond_list(waitfor_deleg
);
757 * mark_caps_dirty - mark some caps dirty
758 * @caps: the dirty caps
760 * note that if there is no dirty and flushing caps before, we need to pin this inode.
761 * it will be unpined by handle_cap_flush_ack when there are no dirty and flushing caps.
763 void Inode::mark_caps_dirty(int caps
)
765 lsubdout(client
->cct
, client
, 10) << __func__
<< " " << *this << " " << ccap_string(dirty_caps
) << " -> "
766 << ccap_string(dirty_caps
| caps
) << dendl
;
767 if (caps
&& !caps_dirty())
770 client
->get_dirty_list().push_back(&dirty_cap_item
);
774 * mark_caps_clean - only clean the dirty_caps and caller should start flushing the dirty caps.
776 void Inode::mark_caps_clean()
778 lsubdout(client
->cct
, client
, 10) << __func__
<< " " << *this << dendl
;
780 dirty_cap_item
.remove_myself();