]>
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 snapdir_parent
->make_nosnap_relative_path(p
);
100 p
.push_dentry(empty
);
106 * make a filepath suitable for an mds request:
107 * - if we are non-snapped/live, the ino is sufficient, e.g. #1234
108 * - if we are snapped, make filepath relative to first non-snapped parent.
110 void Inode::make_nosnap_relative_path(filepath
& p
)
112 if (snapid
== CEPH_NOSNAP
) {
114 } else if (snapdir_parent
) {
115 snapdir_parent
->make_nosnap_relative_path(p
);
117 p
.push_dentry(empty
);
118 } else if (!dentries
.empty()) {
119 Dentry
*dn
= get_first_parent();
120 ceph_assert(dn
->dir
&& dn
->dir
->parent_inode
);
121 dn
->dir
->parent_inode
->make_nosnap_relative_path(p
);
122 p
.push_dentry(dn
->name
);
128 void Inode::get_open_ref(int mode
)
130 open_by_mode
[mode
]++;
131 break_deleg(!(mode
& CEPH_FILE_MODE_WR
));
134 bool Inode::put_open_ref(int mode
)
136 //cout << "open_by_mode[" << mode << "] " << open_by_mode[mode] << " -> " << (open_by_mode[mode]-1) << std::endl;
137 auto& ref
= open_by_mode
.at(mode
);
138 ceph_assert(ref
> 0);
144 void Inode::get_cap_ref(int cap
)
151 //cout << "inode " << *this << " get " << cap_string(c) << " " << (cap_refs[c]-1) << " -> " << cap_refs[c] << std::endl;
158 int Inode::put_cap_ref(int cap
)
165 if (cap_refs
[c
] <= 0) {
166 lderr(client
->cct
) << "put_cap_ref " << ccap_string(c
) << " went negative on " << *this << dendl
;
167 ceph_assert(cap_refs
[c
] > 0);
169 if (--cap_refs
[c
] == 0)
171 //cout << "inode " << *this << " put " << cap_string(c) << " " << (cap_refs[c]+1) << " -> " << cap_refs[c] << std::endl;
179 bool Inode::is_any_caps()
181 return !caps
.empty() || snap_caps
;
184 bool Inode::cap_is_valid(const Cap
&cap
) const
186 /*cout << "cap_gen " << cap->session-> cap_gen << std::endl
187 << "session gen " << cap->gen << std::endl
188 << "cap expire " << cap->session->cap_ttl << std::endl
189 << "cur time " << ceph_clock_now(cct) << std::endl;*/
190 if ((cap
.session
->cap_gen
<= cap
.gen
)
191 && (ceph_clock_now() < cap
.session
->cap_ttl
)) {
197 int Inode::caps_issued(int *implemented
) const
201 for (const auto &pair
: caps
) {
202 const Cap
&cap
= pair
.second
;
203 if (cap_is_valid(cap
)) {
205 i
|= cap
.implemented
;
208 // exclude caps issued by non-auth MDS, but are been revoking by
209 // the auth MDS. The non-auth MDS should be revoking/exporting
210 // these caps, but the message is delayed.
212 c
&= ~auth_cap
->implemented
| auth_cap
->issued
;
219 void Inode::try_touch_cap(mds_rank_t mds
)
221 auto it
= caps
.find(mds
);
222 if (it
!= caps
.end()) {
228 * caps_issued_mask - check whether we have all of the caps in the mask
229 * @mask: mask to check against
230 * @allow_impl: whether the caller can also use caps that are implemented but not issued
232 * This is the bog standard "check whether we have the required caps" operation.
233 * Typically, we only check against the capset that is currently "issued".
234 * In other words, we ignore caps that have been revoked but not yet released.
236 * Some callers (particularly those doing attribute retrieval) can also make
237 * use of the full set of "implemented" caps to satisfy requests from the
240 * Those callers should refrain from taking new references to implemented
243 bool Inode::caps_issued_mask(unsigned mask
, bool allow_impl
)
248 if ((c
& mask
) == mask
)
252 cap_is_valid(*auth_cap
) &&
253 (auth_cap
->issued
& mask
) == mask
) {
258 for (auto &pair
: caps
) {
259 Cap
&cap
= pair
.second
;
260 if (cap_is_valid(cap
)) {
261 if ((cap
.issued
& mask
) == mask
) {
266 i
|= cap
.implemented
;
273 if ((c
& mask
) == mask
) {
274 // bah.. touch them all
275 for (auto &pair
: caps
) {
283 int Inode::caps_used()
286 for (map
<int,int>::iterator p
= cap_refs
.begin();
294 int Inode::caps_file_wanted()
297 for (map
<int,int>::iterator p
= open_by_mode
.begin();
298 p
!= open_by_mode
.end();
301 want
|= ceph_caps_for_mode(p
->first
);
305 int Inode::caps_wanted()
307 int want
= caps_file_wanted() | caps_used();
308 if (want
& CEPH_CAP_FILE_BUFFER
)
309 want
|= CEPH_CAP_FILE_EXCL
;
313 int Inode::caps_mds_wanted()
316 for (const auto &pair
: caps
) {
317 want
|= pair
.second
.wanted
;
322 int Inode::caps_dirty()
324 return dirty_caps
| flushing_caps
;
327 const UserPerm
* Inode::get_best_perms()
329 const UserPerm
*perms
= NULL
;
330 for (const auto &pair
: caps
) {
331 const UserPerm
& iperm
= pair
.second
.latest_perms
;
332 if (!perms
) { // we don't have any, take what's present
334 } else if (iperm
.uid() == uid
) {
335 if (iperm
.gid() == gid
) { // we have the best possible, return
338 if (perms
->uid() != uid
) { // take uid > gid every time
341 } else if (perms
->uid() != uid
&& iperm
.gid() == gid
) {
342 perms
= &iperm
; // a matching gid is better than nothing
348 bool Inode::have_valid_size()
350 // RD+RDCACHE or WR+WRBUFFER => valid size
351 if (caps_issued() & (CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_EXCL
))
356 // open Dir for an inode. if it's not open, allocated it (and pin dentry in memory).
357 Dir
*Inode::open_dir()
361 lsubdout(client
->cct
, client
, 15) << "open_dir " << dir
<< " on " << this << dendl
;
362 ceph_assert(dentries
.size() < 2); // dirs can't be hard-linked
363 if (!dentries
.empty())
364 get_first_parent()->get(); // pin dentry
370 bool Inode::check_mode(const UserPerm
& perms
, unsigned want
)
372 if (uid
== perms
.uid()) {
373 // if uid is owner, owner entry determines access
375 } else if (perms
.gid_in_groups(gid
)) {
376 // if a gid or sgid matches the owning group, group entry determines access
380 return (mode
& want
) == want
;
385 lsubdout(client
->cct
, client
, 15) << "inode.get on " << this << " " << ino
<< '.' << snapid
386 << " now " << _ref
<< dendl
;
389 //private method to put a reference; see Client::put_inode()
390 int Inode::_put(int n
) {
392 lsubdout(client
->cct
, client
, 15) << "inode.put on " << this << " " << ino
<< '.' << snapid
393 << " now " << _ref
<< dendl
;
394 ceph_assert(_ref
>= 0);
399 void Inode::dump(Formatter
*f
) const
401 f
->dump_stream("ino") << ino
;
402 f
->dump_stream("snapid") << snapid
;
404 f
->dump_unsigned("rdev", rdev
);
405 f
->dump_stream("ctime") << ctime
;
406 f
->dump_stream("btime") << btime
;
407 f
->dump_stream("mode") << '0' << std::oct
<< mode
<< std::dec
;
408 f
->dump_unsigned("uid", uid
);
409 f
->dump_unsigned("gid", gid
);
410 f
->dump_int("nlink", nlink
);
412 f
->dump_unsigned("size", size
);
413 f
->dump_unsigned("max_size", max_size
);
414 f
->dump_unsigned("truncate_seq", truncate_seq
);
415 f
->dump_unsigned("truncate_size", truncate_size
);
416 f
->dump_stream("mtime") << mtime
;
417 f
->dump_stream("atime") << atime
;
418 f
->dump_unsigned("time_warp_seq", time_warp_seq
);
419 f
->dump_unsigned("change_attr", change_attr
);
421 f
->dump_object("layout", layout
);
423 f
->open_object_section("dir_layout");
424 ::dump(dir_layout
, f
);
427 f
->dump_bool("complete", flags
& I_COMPLETE
);
428 f
->dump_bool("ordered", flags
& I_DIR_ORDERED
);
430 /* FIXME when wip-mds-encoding is merged ***
431 f->open_object_section("dir_stat");
435 f->open_object_section("rstat");
441 f
->dump_unsigned("version", version
);
442 f
->dump_unsigned("xattr_version", xattr_version
);
443 f
->dump_unsigned("flags", flags
);
446 if (!dir_contacts
.empty()) {
447 f
->open_object_section("dir_contants");
448 for (set
<int>::iterator p
= dir_contacts
.begin(); p
!= dir_contacts
.end(); ++p
)
449 f
->dump_int("mds", *p
);
452 f
->dump_int("dir_hashed", (int)dir_hashed
);
453 f
->dump_int("dir_replicated", (int)dir_replicated
);
456 f
->open_array_section("caps");
457 for (const auto &pair
: caps
) {
458 f
->open_object_section("cap");
459 f
->dump_int("mds", pair
.first
);
460 if (&pair
.second
== auth_cap
)
461 f
->dump_int("auth", 1);
467 f
->dump_int("auth_cap", auth_cap
->session
->mds_num
);
469 f
->dump_stream("dirty_caps") << ccap_string(dirty_caps
);
471 f
->dump_stream("flushings_caps") << ccap_string(flushing_caps
);
472 f
->open_object_section("flushing_cap_tid");
473 for (map
<ceph_tid_t
, int>::const_iterator p
= flushing_cap_tids
.begin();
474 p
!= flushing_cap_tids
.end();
476 string
n(ccap_string(p
->second
));
477 f
->dump_unsigned(n
.c_str(), p
->first
);
481 f
->dump_int("shared_gen", shared_gen
);
482 f
->dump_int("cache_gen", cache_gen
);
484 f
->dump_int("snap_caps", snap_caps
);
485 f
->dump_int("snap_cap_refs", snap_cap_refs
);
488 f
->dump_stream("hold_caps_until") << hold_caps_until
;
491 f
->open_object_section("snaprealm");
495 if (!cap_snaps
.empty()) {
496 for (const auto &p
: cap_snaps
) {
497 f
->open_object_section("cap_snap");
498 f
->dump_stream("follows") << p
.first
;
505 if (!open_by_mode
.empty()) {
506 f
->open_array_section("open_by_mode");
507 for (map
<int,int>::const_iterator p
= open_by_mode
.begin(); p
!= open_by_mode
.end(); ++p
) {
508 f
->open_object_section("ref");
509 f
->dump_int("mode", p
->first
);
510 f
->dump_int("refs", p
->second
);
515 if (!cap_refs
.empty()) {
516 f
->open_array_section("cap_refs");
517 for (map
<int,int>::const_iterator p
= cap_refs
.begin(); p
!= cap_refs
.end(); ++p
) {
518 f
->open_object_section("cap_ref");
519 f
->dump_stream("cap") << ccap_string(p
->first
);
520 f
->dump_int("refs", p
->second
);
526 f
->dump_unsigned("reported_size", reported_size
);
527 if (wanted_max_size
!= max_size
)
528 f
->dump_unsigned("wanted_max_size", wanted_max_size
);
529 if (requested_max_size
!= max_size
)
530 f
->dump_unsigned("requested_max_size", requested_max_size
);
532 f
->dump_int("ref", _ref
);
533 f
->dump_int("ll_ref", ll_ref
);
535 if (!dentries
.empty()) {
536 f
->open_array_section("parents");
537 for (const auto &dn
: dentries
) {
538 f
->open_object_section("dentry");
539 f
->dump_stream("dir_ino") << dn
->dir
->parent_inode
->ino
;
540 f
->dump_string("name", dn
->name
);
547 void Cap::dump(Formatter
*f
) const
549 f
->dump_int("mds", session
->mds_num
);
550 f
->dump_stream("ino") << inode
.ino
;
551 f
->dump_unsigned("cap_id", cap_id
);
552 f
->dump_stream("issued") << ccap_string(issued
);
553 if (implemented
!= issued
)
554 f
->dump_stream("implemented") << ccap_string(implemented
);
555 f
->dump_stream("wanted") << ccap_string(wanted
);
556 f
->dump_unsigned("seq", seq
);
557 f
->dump_unsigned("issue_seq", issue_seq
);
558 f
->dump_unsigned("mseq", mseq
);
559 f
->dump_unsigned("gen", gen
);
562 void CapSnap::dump(Formatter
*f
) const
564 f
->dump_stream("ino") << in
->ino
;
565 f
->dump_stream("issued") << ccap_string(issued
);
566 f
->dump_stream("dirty") << ccap_string(dirty
);
567 f
->dump_unsigned("size", size
);
568 f
->dump_stream("ctime") << ctime
;
569 f
->dump_stream("mtime") << mtime
;
570 f
->dump_stream("atime") << atime
;
571 f
->dump_int("time_warp_seq", time_warp_seq
);
572 f
->dump_stream("mode") << '0' << std::oct
<< mode
<< std::dec
;
573 f
->dump_unsigned("uid", uid
);
574 f
->dump_unsigned("gid", gid
);
575 if (!xattrs
.empty()) {
576 f
->open_object_section("xattr_lens");
577 for (map
<string
,bufferptr
>::const_iterator p
= xattrs
.begin(); p
!= xattrs
.end(); ++p
)
578 f
->dump_int(p
->first
.c_str(), p
->second
.length());
581 f
->dump_unsigned("xattr_version", xattr_version
);
582 f
->dump_int("writing", (int)writing
);
583 f
->dump_int("dirty_data", (int)dirty_data
);
584 f
->dump_unsigned("flush_tid", flush_tid
);
587 void Inode::set_async_err(int r
)
589 for (const auto &fh
: fhs
) {
594 bool Inode::has_recalled_deleg()
596 if (delegations
.empty())
599 // Either all delegations are recalled or none are. Just check the first.
600 Delegation
& deleg
= delegations
.front();
601 return deleg
.is_recalled();
604 void Inode::recall_deleg(bool skip_read
)
606 if (delegations
.empty())
610 for (list
<Delegation
>::iterator d
= delegations
.begin();
611 d
!= delegations
.end(); ++d
) {
613 Delegation
& deleg
= *d
;
614 deleg
.recall(skip_read
);
618 bool Inode::delegations_broken(bool skip_read
)
620 if (delegations
.empty()) {
621 lsubdout(client
->cct
, client
, 10) <<
622 __func__
<< ": delegations empty on " << *this << dendl
;
627 Delegation
& deleg
= delegations
.front();
628 lsubdout(client
->cct
, client
, 10) <<
629 __func__
<< ": read delegs only on " << *this << dendl
;
630 if (deleg
.get_type() == CEPH_FILE_MODE_RD
) {
634 lsubdout(client
->cct
, client
, 10) <<
635 __func__
<< ": not broken" << *this << dendl
;
639 void Inode::break_deleg(bool skip_read
)
641 lsubdout(client
->cct
, client
, 10) <<
642 __func__
<< ": breaking delegs on " << *this << dendl
;
644 recall_deleg(skip_read
);
646 while (!delegations_broken(skip_read
))
647 client
->wait_on_list(waitfor_deleg
);
651 * set_deleg: request a delegation on an open Fh
652 * @fh: filehandle on which to acquire it
653 * @type: delegation request type
654 * @cb: delegation recall callback function
655 * @priv: private pointer to be passed to callback
657 * Attempt to acquire a delegation on an open file handle. If there are no
658 * conflicts and we have the right caps, allocate a new delegation, fill it
659 * out and return 0. Return an error if we can't get one for any reason.
661 int Inode::set_deleg(Fh
*fh
, unsigned type
, ceph_deleg_cb_t cb
, void *priv
)
663 lsubdout(client
->cct
, client
, 10) <<
664 __func__
<< ": inode " << *this << dendl
;
667 * 0 deleg timeout means that they haven't been explicitly enabled. Don't
668 * allow it, with an unusual error to make it clear.
670 if (!client
->get_deleg_timeout())
673 // Just say no if we have any recalled delegs still outstanding
674 if (has_recalled_deleg()) {
675 lsubdout(client
->cct
, client
, 10) << __func__
<<
676 ": has_recalled_deleg" << dendl
;
680 // check vs. currently open files on this inode
682 case CEPH_DELEGATION_RD
:
683 if (open_count_for_write()) {
684 lsubdout(client
->cct
, client
, 10) << __func__
<<
685 ": open for write" << dendl
;
689 case CEPH_DELEGATION_WR
:
690 if (open_count() > 1) {
691 lsubdout(client
->cct
, client
, 10) << __func__
<< ": open" << dendl
;
700 * A delegation is essentially a long-held container for cap references that
701 * we delegate to the client until recalled. The caps required depend on the
702 * type of delegation (read vs. rw). This is entirely an opportunistic thing.
703 * If we don't have the necessary caps for the delegation, then we just don't
706 * In principle we could request the caps from the MDS, but a delegation is
707 * usually requested just after an open. If we don't have the necessary caps
708 * already, then it's likely that there is some sort of conflicting access.
710 * In the future, we may need to add a way to have this request caps more
711 * aggressively -- for instance, to handle WANT_DELEGATION for NFSv4.1+.
713 int need
= ceph_deleg_caps_for_type(type
);
714 if (!caps_issued_mask(need
)) {
715 lsubdout(client
->cct
, client
, 10) << __func__
<< ": cap mismatch, have="
716 << ccap_string(caps_issued()) << " need=" << ccap_string(need
) << dendl
;
720 for (list
<Delegation
>::iterator d
= delegations
.begin();
721 d
!= delegations
.end(); ++d
) {
722 Delegation
& deleg
= *d
;
723 if (deleg
.get_fh() == fh
) {
724 deleg
.reinit(type
, cb
, priv
);
729 delegations
.emplace_back(fh
, type
, cb
, priv
);
734 * unset_deleg - remove a delegation that was previously set
735 * @fh: file handle to clear delegation of
737 * Unlink delegation from the Inode (if there is one), put caps and free it.
739 void Inode::unset_deleg(Fh
*fh
)
741 for (list
<Delegation
>::iterator d
= delegations
.begin();
742 d
!= delegations
.end(); ++d
) {
743 Delegation
& deleg
= *d
;
744 if (deleg
.get_fh() == fh
) {
745 delegations
.erase(d
);
746 client
->signal_cond_list(waitfor_deleg
);
753 * mark_caps_dirty - mark some caps dirty
754 * @caps: the dirty caps
756 * note that if there is no dirty and flushing caps before, we need to pin this inode.
757 * it will be unpined by handle_cap_flush_ack when there are no dirty and flushing caps.
759 void Inode::mark_caps_dirty(int caps
)
761 lsubdout(client
->cct
, client
, 10) << __func__
<< " " << *this << " " << ccap_string(dirty_caps
) << " -> "
762 << ccap_string(dirty_caps
| caps
) << dendl
;
763 if (caps
&& !caps_dirty())
766 client
->get_dirty_list().push_back(&dirty_cap_item
);
770 * mark_caps_clean - only clean the dirty_caps and caller should start flushing the dirty caps.
772 void Inode::mark_caps_clean()
774 lsubdout(client
->cct
, client
, 10) << __func__
<< " " << *this << dendl
;
776 dirty_cap_item
.remove_myself();