1 ===============
2 Pathname lookup
3 ===============
4
5 This write-up is based on three articles publi
6
7 - <https://lwn.net/Articles/649115/> Pathname
8 - <https://lwn.net/Articles/649729/> RCU-walk:
9 - <https://lwn.net/Articles/650786/> A walk am
10
11 Written by Neil Brown with help from Al Viro a
12 It has subsequently been updated to reflect ch
13 including:
14
15 - per-directory parallel name lookup.
16 - ``openat2()`` resolution restriction flags.
17
18 Introduction to pathname lookup
19 ===============================
20
21 The most obvious aspect of pathname lookup, wh
22 exploration is needed to discover, is that it
23 many rules, special cases, and implementation
24 combine to confuse the unwary reader. Compute
25 acquainted with such complexity and has tools
26 tool that we will make extensive use of is "di
27 the early parts of the analysis we will divide
28 them until the final part. Well before we get
29 another major division based on the VFS's appr
30 will allow us to review "REF-walk" and "RCU-wa
31 are getting ahead of ourselves. There are som
32 distinctions we need to clarify first.
33
34 There are two sorts of ...
35 --------------------------
36
37 .. _openat: http://man7.org/linux/man-pages/ma
38
39 Pathnames (sometimes "file names"), used to id
40 filesystem, will be familiar to most readers.
41 of elements: "slashes" that are sequences of o
42 characters, and "components" that are sequence
43 non-"``/``" characters. These form two kinds
44 start with slashes are "absolute" and start fr
45 The others are "relative" and start from the c
46 from some other location specified by a file d
47 "``*at()``" system calls such as `openat() <op
48
49 .. _execveat: http://man7.org/linux/man-pages/
50
51 It is tempting to describe the second kind as
52 component, but that isn't always accurate: a p
53 slashes and components, it can be empty, in ot
54 generally forbidden in POSIX, but some of thos
55 in Linux permit it when the ``AT_EMPTY_PATH``
56 example, if you have an open file descriptor o
57 can execute it by calling `execveat() <execvea
58 the file descriptor, an empty path, and the ``
59
60 These paths can be divided into two sections:
61 everything else. The "everything else" is the
62 it must identify a directory that already exis
63 such as ``ENOENT`` or ``ENOTDIR`` will be repo
64
65 The final component is not so simple. Not onl
66 calls interpret it quite differently (e.g. som
67 not), but it might not even exist: neither the
68 pathname that is just slashes have a final com
69 exist, it could be "``.``" or "``..``" which a
70 from other components.
71
72 .. _POSIX: https://pubs.opengroup.org/onlinepu
73
74 If a pathname ends with a slash, such as "``/t
75 tempting to consider that to have an empty fin
76 ways that would lead to correct results, but n
77 particular, ``mkdir()`` and ``rmdir()`` each c
78 by the final component, and they are required
79 ending in "``/``". According to POSIX_:
80
81 A pathname that contains at least one non-<s
82 that ends with one or more trailing <slash>
83 be resolved successfully unless the last pat
84 the trailing <slash> characters names an exi
85 directory entry that is to be created for a
86 after the pathname is resolved.
87
88 The Linux pathname walking code (mostly in ``f
89 all of these issues: breaking the path into co
90 "everything else" quite separately from the fi
91 checking that the trailing slash is not used w
92 permitted. It also addresses the important is
93 access.
94
95 While one process is looking up a pathname, an
96 changes that affect that lookup. One fairly e
97 "a/b" were renamed to "a/c/b" while another pr
98 "a/b/..", that process might successfully reso
99 Most races are much more subtle, and a big par
100 pathname lookup is to prevent them from having
101 of the possible races are seen most clearly in
102 "dcache" and an understanding of that is centr
103 pathname lookup.
104
105 More than just a cache
106 ----------------------
107
108 The "dcache" caches information about names in
109 make them quickly available for lookup. Each
110 "dentry") contains three significant fields: a
111 pointer to a parent dentry, and a pointer to t
112 contains further information about the object
113 the given name. The inode pointer can be ``NU
114 name doesn't exist in the parent. While there
115 dentry of a directory to the dentries of the c
116 not used for pathname lookup, and so will not
117
118 The dcache has a number of uses apart from acc
119 that will be particularly relevant is that it
120 with the mount table that records which filesy
121 What the mount table actually stores is which
122 of which other dentry.
123
124 When considering the dcache, we have another o
125 distinctions: there are two types of filesyste
126
127 Some filesystems ensure that the information i
128 completely accurate (though not necessarily co
129 the VFS to determine if a particular file does
130 without checking with the filesystem, and mean
131 protect the filesystem against certain races a
132 These are typically "local" filesystems such a
133
134 Other filesystems don't provide that guarantee
135 These are typically filesystems that are share
136 whether remote filesystems like NFS and 9P, or
137 like ocfs2 or cephfs. These filesystems allow
138 cached information, and must provide their own
139 awkward races. The VFS can detect these files
140 ``DCACHE_OP_REVALIDATE`` flag being set in the
141
142 REF-walk: simple concurrency management with r
143 ----------------------------------------------
144
145 With all of those divisions carefully classifi
146 looking at the actual process of walking along
147 we will start with the handling of the "everyt
148 pathname, and focus on the "REF-walk" approach
149 management. This code is found in the ``link_
150 you ignore all the places that only run when "
151 (indicating the use of RCU-walk) is set.
152
153 .. _Meet the Lockers: https://lwn.net/Articles
154
155 REF-walk is fairly heavy-handed with locks and
156 as heavy-handed as in the old "big kernel lock
157 afraid of taking a lock when one is needed. I
158 different concurrency controls. A background
159 various primitives is assumed, or can be glean
160 as in `Meet the Lockers`_.
161
162 The locking mechanisms used by REF-walk includ
163
164 dentry->d_lockref
165 ~~~~~~~~~~~~~~~~~
166
167 This uses the lockref primitive to provide bot
168 reference count. The special-sauce of this pr
169 conceptual sequence "lock; inc_ref; unlock;" c
170 with a single atomic memory operation.
171
172 Holding a reference on a dentry ensures that t
173 be freed and used for something else, so the v
174 will behave as expected. It also protects the
175 to the inode to some extent.
176
177 The association between a dentry and its inode
178 For example, when a file is renamed, the dentr
179 together to the new location. When a file is
180 initially be negative (i.e. ``d_inode`` is ``N
181 to the new inode as part of the act of creatio
182
183 When a file is deleted, this can be reflected
184 setting ``d_inode`` to ``NULL``, or by removin
185 (described shortly) used to look up the name i
186 If the dentry is still in use the second optio
187 perfectly legal to keep using an open file aft
188 and having the dentry around helps. If the de
189 use (i.e. if the refcount in ``d_lockref`` is
190 ``d_inode`` be set to ``NULL``. Doing it this
191 very common case.
192
193 So as long as a counted reference is held to a
194 value will never be changed.
195
196 dentry->d_lock
197 ~~~~~~~~~~~~~~
198
199 ``d_lock`` is a synonym for the spinlock that
200 For our purposes, holding this lock protects a
201 renamed or unlinked. In particular, its paren
202 name (``d_name``) cannot be changed, and it ca
203 dentry hash table.
204
205 When looking for a name in a directory, REF-wa
206 each candidate dentry that it finds in the has
207 that the parent and name are correct. So it d
208 while searching in the cache; it only locks ch
209
210 When looking for the parent for a given name (
211 REF-walk can take ``d_lock`` to get a stable r
212 but it first tries a more lightweight approach
213 ``dget_parent()``, if a reference can be claim
214 subsequently ``d_parent`` can be seen to have
215 no need to actually take the lock on the child
216
217 rename_lock
218 ~~~~~~~~~~~
219
220 Looking up a given name in a given directory i
221 from the two values (the name and the dentry o
222 accessing that slot in a hash table, and searc
223 that is found there.
224
225 When a dentry is renamed, the name and the par
226 change so the hash will almost certainly chang
227 dentry to a different chain in the hash table.
228 happened to be looking at a dentry that was mo
229 it might end up continuing the search down the
230 and so miss out on part of the correct chain.
231
232 The name-lookup process (``d_lookup()``) does
233 from happening, but only to detect when it hap
234 ``rename_lock`` is a seqlock that is updated w
235 renamed. If ``d_lookup`` finds that a rename
236 unsuccessfully scanned a chain in the hash tab
237 again.
238
239 ``rename_lock`` is also used to detect and def
240 against ``LOOKUP_BENEATH`` and ``LOOKUP_IN_ROO
241 the parent directory is moved outside the root
242 check). If ``rename_lock`` is updated during t
243 a "..", a potential attack occurred and ``hand
244 ``-EAGAIN``.
245
246 inode->i_rwsem
247 ~~~~~~~~~~~~~~
248
249 ``i_rwsem`` is a read/write semaphore that ser
250 directory. This ensures that, for example, an
251 cannot both happen at the same time. It also
252 stable while the filesystem is asked to look u
253 currently in the dcache or, optionally, when t
254 directory is being retrieved with ``readdir()`
255
256 This has a complementary role to that of ``d_l
257 directory protects all of the names in that di
258 on a name protects just one name in a director
259 dcache hold ``i_rwsem`` on the relevant direct
260 ``d_lock`` on one or more the dentries while t
261 exception is when idle dentries are removed fr
262 memory pressure. This uses ``d_lock``, but ``
263
264 The semaphore affects pathname lookup in two d
265 prevents changes during lookup of a name in a
266 ``lookup_fast()`` first which, in turn, checks
267 using only ``d_lock`` locking. If the name is
268 falls back to ``lookup_slow()`` which takes a
269 the name isn't in the cache, and then calls in
270 definitive answer. A new dentry will be added
271 the result.
272
273 Secondly, when pathname lookup reaches the fin
274 sometimes need to take an exclusive lock on ``
275 that the required exclusion can be achieved.
276 to take, or not take, ``i_rwsem`` is one of th
277 issues addressed in a subsequent section.
278
279 If two threads attempt to look up the same nam
280 name that is not yet in the dcache - the share
281 not prevent them both adding new dentries with
282 would result in confusion an extra level of in
283 based around a secondary hash table (``in_look
284 per-dentry flag bit (``DCACHE_PAR_LOOKUP``).
285
286 To add a new dentry to the cache while only ho
287 ``i_rwsem``, a thread must call ``d_alloc_para
288 dentry, stores the required name and parent in
289 is already a matching dentry in the primary or
290 tables, and if not, stores the newly allocated
291 hash table, with ``DCACHE_PAR_LOOKUP`` set.
292
293 If a matching dentry was found in the primary
294 returned and the caller can know that it lost
295 thread adding the entry. If no matching dentr
296 cache, the newly allocated dentry is returned
297 detect this from the presence of ``DCACHE_PAR_
298 knows that it has won any race and now is resp
299 filesystem to perform the lookup and find the
300 the lookup is complete, it must call ``d_looku
301 the flag and does some other house keeping, in
302 dentry from the secondary hash table - it will
303 added to the primary hash table already. Note
304 waitqueue_head`` is passed to ``d_alloc_parall
305 ``d_lookup_done()`` must be called while this
306 in scope.
307
308 If a matching dentry is found in the secondary
309 ``d_alloc_parallel()`` has a little more work
310 ``DCACHE_PAR_LOOKUP`` to be cleared, using a w
311 to the instance of ``d_alloc_parallel()`` that
312 will be woken by the call to ``d_lookup_done()
313 if the dentry has now been added to the primar
314 has, the dentry is returned and the caller jus
315 race. If it hasn't been added to the primary
316 likely explanation is that some other dentry w
317 ``d_splice_alias()``. In any case, ``d_alloc_
318 look ups from the start and will normally retu
319 primary hash table.
320
321 mnt->mnt_count
322 ~~~~~~~~~~~~~~
323
324 ``mnt_count`` is a per-CPU reference counter o
325 Per-CPU here means that incrementing the count
326 uses CPU-local memory, but checking if the cou
327 it needs to check with every CPU. Taking a ``
328 prevents the mount structure from disappearing
329 unmount operations, but does not prevent a "la
330 ``mnt_count`` doesn't ensure that the mount re
331 in particular, doesn't stabilize the link to t
332 does, however, ensure that the ``mount`` data
333 and it provides a reference to the root dentry
334 filesystem. So a reference through ``->mnt_co
335 reference to the mounted dentry, but not the m
336
337 mount_lock
338 ~~~~~~~~~~
339
340 ``mount_lock`` is a global seqlock, a bit like
341 check if any change has been made to any mount
342
343 While walking down the tree (away from the roo
344 crossing a mount point to check that the cross
345 the value in the seqlock is read, then the cod
346 is mounted on the current directory, if there
347 the ``mnt_count``. Finally the value in ``mou
348 the old value. If there is no change, then th
349 was a change, the ``mnt_count`` is decremented
350 retried.
351
352 When walking up the tree (towards the root) by
353 a little more care is needed. In this case th
354 contains both a counter and a spinlock) is ful
355 any changes to any mount points while stepping
356 needed to stabilize the link to the mounted-on
357 refcount on the mount itself doesn't ensure.
358
359 ``mount_lock`` is also used to detect and defe
360 against ``LOOKUP_BENEATH`` and ``LOOKUP_IN_ROO
361 the parent directory is moved outside the root
362 check). If ``mount_lock`` is updated during th
363 a "..", a potential attack occurred and ``hand
364 ``-EAGAIN``.
365
366 RCU
367 ~~~
368
369 Finally the global (but extremely lightweight)
370 from time to time to ensure certain data struc
371 unexpectedly.
372
373 In particular it is held while scanning chains
374 table, and the mount point hash table.
375
376 Bringing it together with ``struct nameidata``
377 ----------------------------------------------
378
379 .. _First edition Unix: https://minnie.tuhs.or
380
381 Throughout the process of walking a path, the
382 in a ``struct nameidata``, "namei" being the t
383 all the way back to `First Edition Unix`_ - of
384 converts a "name" to an "inode". ``struct nam
385 other fields):
386
387 ``struct path path``
388 ~~~~~~~~~~~~~~~~~~~~
389
390 A ``path`` contains a ``struct vfsmount`` (whi
391 embedded in a ``struct mount``) and a ``struct
392 record the current status of the walk. They s
393 starting point (the current working directory,
394 directory identified by a file descriptor), an
395 step. A reference through ``d_lockref`` and `
396 held.
397
398 ``struct qstr last``
399 ~~~~~~~~~~~~~~~~~~~~
400
401 This is a string together with a length (i.e.
402 that is the "next" component in the pathname.
403
404 ``int last_type``
405 ~~~~~~~~~~~~~~~~~
406
407 This is one of ``LAST_NORM``, ``LAST_ROOT``, `
408 The ``last`` field is only valid if the type i
409
410 ``struct path root``
411 ~~~~~~~~~~~~~~~~~~~~
412
413 This is used to hold a reference to the effect
414 filesystem. Often that reference won't be nee
415 only assigned the first time it is used, or wh
416 is requested. Keeping a reference in the ``na
417 only one root is in effect for the entire path
418 with a ``chroot()`` system call.
419
420 It should be noted that in the case of ``LOOKU
421 ``LOOKUP_BENEATH``, the effective root becomes
422 passed to ``openat2()`` (which exposes these `
423
424 The root is needed when either of two conditio
425 pathname or a symbolic link starts with a "'/'
426 component is being handled, since "``..``" fro
427 at the root. The value used is usually the cu
428 the calling process. An alternate root can be
429 ``sysctl()`` calls ``file_open_root()``, and w
430 ``mount_subtree()``. In each case a pathname
431 specific part of the filesystem, and the looku
432 escape that subtree. It works a bit like a lo
433
434 Ignoring the handling of symbolic links, we ca
435 "``link_path_walk()``" function, which handles
436 except the final component as:
437
438 Given a path (``name``) and a nameidata str
439 current directory has execute permission an
440 over one component while updating ``last_ty
441 was the final component, then return, other
442 ``walk_component()`` and repeat from the to
443
444 ``walk_component()`` is even easier. If the c
445 it calls ``handle_dots()`` which does the nece
446 described. If it finds a ``LAST_NORM`` compon
447 "``lookup_fast()``" which only looks in the dc
448 filesystem to revalidate the result if it is t
449 If that doesn't get a good result, it calls "`
450 takes ``i_rwsem``, rechecks the cache, and the
451 to find a definitive answer.
452
453 As the last step of walk_component(), step_int
454 directly from walk_component() or from handle_
455 handle_mounts(), to check and handle mount poi
456 ``struct path`` is created containing a counte
457 a reference to the new ``vfsmount`` which is o
458 different from the previous ``vfsmount``. Then
459 a symbolic link, step_into() calls pick_link()
460 otherwise it installs the new ``struct path``
461 drops the unneeded references.
462
463 This "hand-over-hand" sequencing of getting a
464 dentry before dropping the reference to the pr
465 seem obvious, but is worth pointing out so tha
466 analogue in the "RCU-walk" version.
467
468 Handling the final component
469 ----------------------------
470
471 ``link_path_walk()`` only walks as far as sett
472 ``nd->last_type`` to refer to the final compon
473 not call ``walk_component()`` that last time.
474 component remains for the caller to sort out.
475 path_lookupat(), path_parentat() and
476 path_openat() each of which handles the differ
477 different system calls.
478
479 ``path_parentat()`` is clearly the simplest -
480 of housekeeping around ``link_path_walk()`` an
481 directory and final component to the caller.
482 aiming to create a name (via ``filename_create
483 a name (in which case ``user_path_parent()`` i
484 ``i_rwsem`` to exclude other changes while the
485 perform their operation.
486
487 ``path_lookupat()`` is nearly as simple - it i
488 object is wanted such as by ``stat()`` or ``ch
489 calls ``walk_component()`` on the final compon
490 ``lookup_last()``. ``path_lookupat()`` return
491 It is worth noting that when flag ``LOOKUP_MOU
492 path_lookupat() will unset LOOKUP_JUMPED in na
493 subsequent path traversal d_weak_revalidate()
494 This is important when unmounting a filesystem
495 one provided by a dead NFS server.
496
497 Finally ``path_openat()`` is used for the ``op
498 contains, in support functions starting with "
499 complexity needed to handle the different subt
500 or without O_EXCL), final "``/``" characters,
501 links. We will revisit this in the final part
502 focuses on those symbolic links. "open_last_l
503 not always, take ``i_rwsem``, depending on wha
504
505 Each of these, or the functions which call the
506 the possibility that the final component is no
507 goal of the lookup is to create something, the
508 ``last_type`` other than ``LAST_NORM`` will re
509 example if ``path_parentat()`` reports ``LAST_
510 won't try to create that name. They also chec
511 by testing ``last.name[last.len]``. If there
512 the final component, it must be a trailing sla
513
514 Revalidation and automounts
515 ---------------------------
516
517 Apart from symbolic links, there are only two
518 process not yet covered. One is the handling
519 and the other is automounts.
520
521 On filesystems that require it, the lookup rou
522 ``->d_revalidate()`` dentry method to ensure t
523 is current. This will often confirm validity
524 from a server. In some cases it may find that
525 further up the path and that something that wa
526 previously isn't really. When this happens th
527 path is aborted and retried with the "``LOOKUP
528 forces revalidation to be more thorough. We w
529 this retry process in the next article.
530
531 Automount points are locations in the filesyst
532 lookup a name can trigger changes to how that
533 handled, in particular by mounting a filesyste
534 covered in greater detail in autofs.txt in the
535 tree, but a few notes specifically related to
536 here.
537
538 The Linux VFS has a concept of "managed" dentr
539 potentially interesting things about these den
540 to three different flags that might be set in
541
542 ``DCACHE_MANAGE_TRANSIT``
543 ~~~~~~~~~~~~~~~~~~~~~~~~~
544
545 If this flag has been set, then the filesystem
546 ``d_manage()`` dentry operation be called befo
547 mount point. This can perform two particular
548
549 It can block to avoid races. If an automount
550 unmounted, the ``d_manage()`` function will us
551 process to complete before letting the new loo
552 trigger a new automount.
553
554 It can selectively allow only some processes t
555 mount point. When a server process is managin
556 need to access a directory without triggering
557 processing. That server process can identify
558 filesystem, which will then give it a special
559 ``d_manage()`` by returning ``-EISDIR``.
560
561 ``DCACHE_MOUNTED``
562 ~~~~~~~~~~~~~~~~~~
563
564 This flag is set on every dentry that is mount
565 supports multiple filesystem namespaces, it is
566 dentry may not be mounted on in *this* namespa
567 other. So this flag is seen as a hint, not a
568
569 If this flag is set, and ``d_manage()`` didn't
570 ``lookup_mnt()`` is called to examine the moun
571 ``mount_lock`` described earlier) and possibly
572 and a new ``dentry`` (both with counted refere
573
574 ``DCACHE_NEED_AUTOMOUNT``
575 ~~~~~~~~~~~~~~~~~~~~~~~~~
576
577 If ``d_manage()`` allowed us to get this far,
578 find a mount point, then this flag causes the
579 operation to be called.
580
581 The ``d_automount()`` operation can be arbitra
582 communicate with server processes etc. but it
583 report that there was an error, that there was
584 should provide an updated ``struct path`` with
585
586 In the latter case, ``finish_automount()`` wil
587 install the new mount point into the mount tab
588
589 There is no new locking of import here and it
590 locks (only counted references) are held over
591 the very real possibility of extended delays.
592 This will become more important next time when
593 which is particularly sensitive to delays.
594
595 RCU-walk - faster pathname lookup in Linux
596 ==========================================
597
598 RCU-walk is another algorithm for performing p
599 It is in many ways similar to REF-walk and the
600 of code. The significant difference in RCU-wa
601 the possibility of concurrent access.
602
603 We noted that REF-walk is complex because ther
604 and special cases. RCU-walk reduces this comp
605 refusing to handle a number of cases -- it ins
606 REF-walk. The difficulty with RCU-walk comes
607 direction: unfamiliarity. The locking rules w
608 quite different from traditional locking, so w
609 time when we come to those.
610
611 Clear demarcation of roles
612 --------------------------
613
614 The easiest way to manage concurrency is to fo
615 thread from changing the data structures that
616 looking at. In cases where no other thread wo
617 changing the data and lots of different thread
618 same time, this can be very costly. Even when
619 multiple concurrent readers, the simple act of
620 the number of current readers can impose an un
621 goal when reading a shared data structure that
622 changing is to avoid writing anything to memor
623 locks, increment no counts, leave no footprint
624
625 The REF-walk mechanism already described certa
626 principle, but then it is really designed to w
627 be other threads modifying the data. RCU-walk
628 designed for the common situation where there
629 readers and only occasional writers. This may
630 parts of the filesystem tree, but in many part
631 other parts it is important that RCU-walk can
632 using REF-walk.
633
634 Pathname lookup always starts in RCU-walk mode
635 as long as what it is looking for is in the ca
636 dances lightly down the cached filesystem imag
637 and carefully watching where it is, to be sure
638 notices that something has changed or is chang
639 isn't in the cache, then it tries to stop grac
640 REF-walk.
641
642 This stopping requires getting a counted refer
643 ``vfsmount`` and ``dentry``, and ensuring that
644 that a path walk with REF-walk would have foun
645 This is an invariant that RCU-walk must guaran
646 decisions, such as selecting the next step, th
647 REF-walk could also have made if it were walki
648 same time. If the graceful stop succeeds, the
649 processed with the reliable, if slightly slugg
650 RCU-walk finds it cannot stop gracefully, it s
651 restarts from the top with REF-walk.
652
653 This pattern of "try RCU-walk, if that fails t
654 clearly seen in functions like filename_lookup
655 filename_parentat(),
656 do_filp_open(), and do_file_open_root(). Thes
657 correspond roughly to the three ``path_*()`` f
658 each of which calls ``link_path_walk()``. The
659 called using different mode flags until a mode
660 They are first called with ``LOOKUP_RCU`` set
661 that fails with the error ``ECHILD`` they are
662 special flag to request "REF-walk". If either
663 error ``ESTALE`` a final attempt is made with
664 ``LOOKUP_RCU``) to ensure that entries found i
665 revalidated - normally entries are only revali
666 determines that they are too old to trust.
667
668 The ``LOOKUP_RCU`` attempt may drop that flag
669 REF-walk, but will never then try to switch ba
670 that trip up RCU-walk are much more likely to
671 so it is very unlikely that there will be much
672 switching back.
673
674 RCU and seqlocks: fast and light
675 --------------------------------
676
677 RCU is, unsurprisingly, critical to RCU-walk m
678 ``rcu_read_lock()`` is held for the entire tim
679 down a path. The particular guarantee it prov
680 data structures - dentries, inodes, super_bloc
681 not be freed while the lock is held. They mig
682 invalidated in one way or another, but the mem
683 repurposed so values in various fields will st
684 is the only guarantee that RCU provides; every
685 seqlocks.
686
687 As we saw above, REF-walk holds a counted refe
688 dentry and the current vfsmount, and does not
689 before taking references to the "next" dentry
690 sometimes takes the ``d_lock`` spinlock. Thes
691 taken to prevent certain changes from happenin
692 take those references or locks and so cannot p
693 Instead, it checks to see if a change has been
694 retries if it has.
695
696 To preserve the invariant mentioned above (tha
697 decisions that REF-walk could have made), it m
698 or near the same places that REF-walk holds th
699 REF-walk increments a reference count or takes
700 samples the status of a seqlock using ``read_s
701 similar function. When REF-walk decrements th
702 lock, RCU-walk checks if the sampled status is
703 ``read_seqcount_retry()`` or similar.
704
705 However, there is a little bit more to seqlock
706 RCU-walk accesses two different fields in a se
707 structure, or accesses the same field twice, t
708 guarantee of any consistency between those acc
709 is needed - which it usually is - RCU-walk mus
710 use ``read_seqcount_retry()`` to validate that
711
712 ``read_seqcount_retry()`` not only checks the
713 imposes a memory barrier so that no memory-rea
714 *before* the call can be delayed until *after*
715 CPU or by the compiler. A simple example of t
716 ``slow_dentry_cmp()`` which, for filesystems w
717 byte-wise name equality, calls into the filesy
718 against a dentry. The length and name pointer
719 variables, then ``read_seqcount_retry()`` is c
720 are consistent, and only then is ``->d_compare
721 standard filename comparison is used, ``dentry
722 instead. Notably it does *not* use ``read_seq
723 instead has a large comment explaining why the
724 isn't necessary. A subsequent ``read_seqcount
725 sufficient to catch any problem that could occ
726
727 With that little refresher on seqlocks out of
728 the bigger picture of how RCU-walk uses seqloc
729
730 ``mount_lock`` and ``nd->m_seq``
731 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
732
733 We already met the ``mount_lock`` seqlock when
734 ensure that crossing a mount point is performe
735 it for that too, but for quite a bit more.
736
737 Instead of taking a counted reference to each
738 descends the tree, RCU-walk samples the state
739 start of the walk and stores this initial sequ
740 ``struct nameidata`` in the ``m_seq`` field.
741 sequence number are used to validate all acces
742 and all mount point crossings. As changes to
743 relatively rare, it is reasonable to fall back
744 that any "mount" or "unmount" happens.
745
746 ``m_seq`` is checked (using ``read_seqretry()`
747 sequence, whether switching to REF-walk for th
748 when the end of the path is reached. It is al
749 down over a mount point (in ``__follow_mount_r
750 ``follow_dotdot_rcu()``). If it is ever found
751 whole RCU-walk sequence is aborted and the pat
752 REF-walk.
753
754 If RCU-walk finds that ``mount_lock`` hasn't c
755 that, had REF-walk taken counted references on
756 results would have been the same. This ensure
757 at least for vfsmount structures.
758
759 ``dentry->d_seq`` and ``nd->seq``
760 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
761
762 In place of taking a count or lock on ``d_refl
763 the per-dentry ``d_seq`` seqlock, and stores t
764 ``seq`` field of the nameidata structure, so `
765 the current sequence number of ``nd->dentry``.
766 revalidated after copying, and before using, t
767 inode of the dentry.
768
769 The handling of the name we have already looke
770 only accessed in ``follow_dotdot_rcu()`` which
771 the required pattern, though it does so for th
772
773 When not at a mount point, ``d_parent`` is fol
774 collected. When we are at a mount point, we i
775 ``mnt->mnt_mountpoint`` link to get a new dent
776 ``d_seq``. Then, after finally finding a ``d_
777 check if we have landed on a mount point and,
778 mount point and follow the ``mnt->mnt_root`` l
779 somewhat unusual, but certainly possible, circ
780 starting point of the path lookup was in part
781 was mounted on, and so not visible from the ro
782
783 The inode pointer, stored in ``->d_inode``, is
784 interesting. The inode will always need to be
785 twice, once to determine if it is NULL and onc
786 permissions. Symlink handling requires a vali
787 Rather than revalidating on each access, a cop
788 access and it is stored in the ``inode`` field
789 it can be safely accessed without further vali
790
791 ``lookup_fast()`` is the only lookup routine t
792 ``lookup_slow()`` being too slow and requiring
793 ``lookup_fast()`` that we find the important "
794 of the current dentry.
795
796 The current ``dentry`` and current ``seq`` num
797 ``__d_lookup_rcu()`` which, on success, return
798 new ``seq`` number. ``lookup_fast()`` then co
799 revalidates the new ``seq`` number. It then v
800 with the old ``seq`` number one last time and
801 process of getting the ``seq`` number of the n
802 checking the ``seq`` number of the old exactly
803 getting a counted reference to the new dentry
804 the old dentry which we saw in REF-walk.
805
806 No ``inode->i_rwsem`` or even ``rename_lock``
807 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
808
809 A semaphore is a fairly heavyweight lock that
810 permissible to sleep. As ``rcu_read_lock()``
811 ``inode->i_rwsem`` plays no role in RCU-walk.
812 take ``i_rwsem`` and modifies the directory in
813 to notice, the result will be either that RCU-
814 dentry that it is looking for, or it will find
815 ``read_seqretry()`` won't validate. In either
816 REF-walk mode which can take whatever locks ar
817
818 Though ``rename_lock`` could be used by RCU-wa
819 any sleeping, RCU-walk doesn't bother. REF-wa
820 protect against the possibility of hash chains
821 while they are being searched. This can resul
822 something that actually is there. When RCU-wa
823 something in the dentry cache, whether it is r
824 already drops down to REF-walk and tries again
825 locking. This neatly handles all cases, so ad
826 rename_lock would bring no significant value.
827
828 ``unlazy walk()`` and ``complete_walk()``
829 -----------------------------------------
830
831 That "dropping down to REF-walk" typically inv
832 ``unlazy_walk()``, so named because "RCU-walk"
833 referred to as "lazy walk". ``unlazy_walk()``
834 following the path down to the current vfsmoun
835 have proceeded successfully, but the next step
836 can happen if the next name cannot be found in
837 permission checking or name revalidation could
838 the ``rcu_read_lock()`` is held (which forbids
839 automount point is found, or in a couple of ca
840 It is also called from ``complete_walk()`` whe
841 the final component, or the very end of the pa
842 particular flavor of lookup is used.
843
844 Other reasons for dropping out of RCU-walk tha
845 to ``unlazy_walk()`` are when some inconsisten
846 handled immediately, such as ``mount_lock`` or
847 seqlocks reporting a change. In these cases t
848 will return ``-ECHILD`` which will percolate u
849 attempt from the top using REF-walk.
850
851 For those cases where ``unlazy_walk()`` is an
852 takes a reference on each of the pointers that
853 dentry, and possibly some symbolic links) and
854 relevant seqlocks have not been changed. If t
855 it, too, aborts with ``-ECHILD``, otherwise th
856 has been a success and the lookup process cont
857
858 Taking a reference on those pointers is not qu
859 incrementing a counter. That works to take a
860 already have one (often indirectly through ano
861 isn't sufficient if you don't actually have a
862 all. For ``dentry->d_lockref``, it is safe to
863 counter to get a reference unless it has been
864 "dead" which involves setting the counter to `
865 ``lockref_get_not_dead()`` achieves this.
866
867 For ``mnt->mnt_count`` it is safe to take a re
868 ``mount_lock`` is then used to validate the re
869 validation fails, it may *not* be safe to just
870 the standard way of calling ``mnt_put()`` - an
871 progressed too far. So the code in ``legitimi
872 finds that the reference it got might not be s
873 ``MNT_SYNC_UMOUNT`` flag to determine if a sim
874 correct, or if it should just decrement the co
875 this ever happened.
876
877 Taking care in filesystems
878 --------------------------
879
880 RCU-walk depends almost entirely on cached inf
881 not call into the filesystem at all. However
882 besides the already-mentioned component-name c
883 file system might be included in RCU-walk, and
884 careful.
885
886 If the filesystem has non-standard permission-
887 such as a networked filesystem which may need
888 - the ``i_op->permission`` interface might be
889 In this case an extra "``MAY_NOT_BLOCK``" flag
890 knows not to sleep, but to return ``-ECHILD``
891 promptly. ``i_op->permission`` is given the i
892 dentry, so it doesn't need to worry about furt
893 However if it accesses any other filesystem da
894 ensure they are safe to be accessed with only
895 held. This typically means they must be freed
896 similar.
897
898 .. _READ_ONCE: https://lwn.net/Articles/624126
899
900 If the filesystem may need to revalidate dcach
901 ``d_op->d_revalidate`` may be called in RCU-wa
902 *is* passed the dentry but does not have acces
903 ``seq`` number from the ``nameidata``, so it n
904 when accessing fields in the dentry. This "ex
905 involves using `READ_ONCE() <READ_ONCE_>`_ to
906 result is not NULL before using it. This patt
907 ``nfs_lookup_revalidate()``.
908
909 A pair of patterns
910 ------------------
911
912 In various places in the details of REF-walk a
913 the big picture, there are a couple of related
914 being aware of.
915
916 The first is "try quickly and check, if that f
917 can see that in the high-level approach of fir
918 then trying REF-walk, and in places where ``un
919 switch to REF-walk for the rest of the path.
920 in ``dget_parent()`` when following a "``..``"
921 to get a reference, then falls back to taking
922
923 The second pattern is "try quickly and check,
924 again - repeatedly". This is seen with the us
925 ``mount_lock`` in REF-walk. RCU-walk doesn't
926 if anything goes wrong it is much safer to jus
927 sedate approach.
928
929 The emphasis here is "try quickly and check".
930 "try quickly *and carefully*, then check". Th
931 needed is a reminder that the system is dynami
932 number of things are safe at all. The most li
933 this whole process is assuming something is sa
934 isn't. Careful consideration of what exactly
935 each access is sometimes necessary.
936
937 A walk among the symlinks
938 =========================
939
940 There are several basic issues that we will ex
941 handling of symbolic links: the symlink stack
942 lifetimes, will help us understand the overall
943 symlinks and lead to the special care needed f
944 Then a consideration of access-time updates an
945 flags controlling lookup will finish the story
946
947 The symlink stack
948 -----------------
949
950 There are only two sorts of filesystem objects
951 appear in a path prior to the final component:
952 Handling directories is quite straightforward:
953 simply becomes the starting point at which to
954 component on the path. Handling symbolic link
955 work.
956
957 Conceptually, symbolic links could be handled
958 a component name refers to a symbolic link, th
959 replaced by the body of the link and, if that
960 then all preceding parts of the path are disca
961 "``readlink -f``" command does, though it also
962 "``..``" components.
963
964 Directly editing the path string is not really
965 up a path, and discarding early components is
966 looked at anyway. Keeping track of all remain
967 important, but they can of course be kept sepa
968 to concatenate them. As one symlink may easil
969 which in turn can refer to a third, we may nee
970 components of several paths, each to be proces
971 ones are completed. These path remnants are k
972 limited size.
973
974 There are two reasons for placing limits on ho
975 occur in a single path lookup. The most obvio
976 If a symlink referred to itself either directl
977 intermediaries, then following the symlink can
978 successfully - the error ``ELOOP`` must be ret
979 detected without imposing limits, but limits a
980 and, given the second reason for restriction,
981
982 .. _outlined recently: http://thread.gmane.org
983
984 The second reason was `outlined recently`_ by
985
986 Because it's a latency and DoS issue too. W
987 true loops, but also to "very deep" non-loo
988 use, it's about users triggering unreasonab
989
990 Linux imposes a limit on the length of any pat
991 is 4096. There are a number of reasons for th
992 kernel spend too much time on just one path is
993 symbolic links you can effectively generate mu
994 sort of limit is needed for the same reason.
995 at most 40 (MAXSYMLINKS) symlinks in any one p
996 a further limit of eight on the maximum depth
997 raised to 40 when a separate stack was impleme
998 just the one limit.
999
1000 The ``nameidata`` structure that we met in an
1001 small stack that can be used to store the rem
1002 symlinks. In many cases this will be suffici
1003 separate stack is allocated with room for 40
1004 lookup will never exceed that stack as, once
1005 detected, an error is returned.
1006
1007 It might seem that the name remnants are all
1008 this stack, but we need a bit more. To see t
1009 cache lifetimes.
1010
1011 Storage and lifetime of cached symlinks
1012 ---------------------------------------
1013
1014 Like other filesystem resources, such as inod
1015 entries, symlinks are cached by Linux to avoi
1016 to external storage. It is particularly impo
1017 able to find and temporarily hold onto these
1018 it doesn't need to drop down into REF-walk.
1019
1020 .. _object-oriented design pattern: https://l
1021
1022 While each filesystem is free to make its own
1023 typically stored in one of two places. Short
1024 stored directly in the inode. When a filesys
1025 inode`` it typically allocates extra space to
1026 common `object-oriented design pattern`_ in t
1027 sometimes include space for a symlink. The o
1028 in the page cache, which normally stores the
1029 pathname in a symlink can be seen as the cont
1030 can easily be stored in the page cache just l
1031
1032 When neither of these is suitable, the next m
1033 that the filesystem will allocate some tempor
1034 construct the symlink content into that memor
1035
1036 When the symlink is stored in the inode, it h
1037 the inode which, itself, is protected by RCU
1038 on the dentry. This means that the mechanism
1039 uses to access the dcache and icache (inode c
1040 sufficient for accessing some cached symlinks
1041 the ``i_link`` pointer in the inode is set to
1042 symlink is stored and it can be accessed dire
1043
1044 When the symlink is stored in the page cache
1045 situation is not so straightforward. A refer
1046 on an inode does not imply any reference on c
1047 inode, and even an ``rcu_read_lock()`` is not
1048 a page will not disappear. So for these syml
1049 code needs to ask the filesystem to provide a
1050 significantly, needs to release that referenc
1051 with it.
1052
1053 Taking a reference to a cache page is often p
1054 mode. It does require making changes to memo
1055 but that isn't necessarily a big cost and it
1056 out of RCU-walk mode completely. Even filesy
1057 space to copy the symlink into can use ``GFP_
1058 allocate memory without the need to drop out
1059 filesystem cannot successfully get a referenc
1060 must return ``-ECHILD`` and ``unlazy_walk()``
1061 REF-walk mode in which the filesystem is allo
1062
1063 The place for all this to happen is the ``i_o
1064 method. This is called both in RCU-walk and R
1065 ``dentry*`` argument is NULL, ``->get_link()`
1066 RCU-walk. Much like the ``i_op->permission()
1067 looked at previously, ``->get_link()`` would
1068 all the data structures it references are saf
1069 holding no counted reference, only the RCU lo
1070 ``struct delayed_called`` will be passed to `
1071 file systems can set their own put_link funct
1072 set_delayed_call(). Later on, when VFS wants
1073 do_delayed_call() to invoke that callback fun
1074
1075 In order for the reference to each symlink to
1076 whether in RCU-walk or REF-walk, the symlink
1077 along with the path remnants:
1078
1079 - the ``struct path`` to provide a reference
1080 - the ``const char *`` to provide a reference
1081 - the ``seq`` to allow the path to be safely
1082 - the ``struct delayed_call`` for later invoc
1083
1084 This means that each entry in the symlink sta
1085 pointers and an integer instead of just one p
1086 remnant). On a 64-bit system, this is about
1087 with 40 entries it adds up to 1600 bytes tota
1088 half a page. So it might seem like a lot, bu
1089 excessive.
1090
1091 Note that, in a given stack frame, the path r
1092 part of the symlink that the other fields ref
1093 to be followed once that symlink has been ful
1094
1095 Following the symlink
1096 ---------------------
1097
1098 The main loop in ``link_path_walk()`` iterate
1099 components in the path and all of the non-fin
1100 are processed, the ``name`` pointer is adjust
1101 symlink, or is restored from the stack, so th
1102 doesn't need to notice. Getting this ``name`
1103 stack is very straightforward; pushing and po
1104 a little more complex.
1105
1106 When a symlink is found, walk_component() cal
1107 which returns the link from the filesystem.
1108 Providing that operation is successful, the o
1109 stack, and the new value is used as the ``nam
1110 the path is found (i.e. ``*name`` is ``'\0'``
1111 off the stack and path walking continues.
1112
1113 Pushing and popping the reference pointers (i
1114 complex in part because of the desire to hand
1115 the last component of a symlink itself points
1116 want to pop the symlink-just-completed off th
1117 the symlink-just-found to avoid leaving empty
1118 just get in the way.
1119
1120 It is most convenient to push the new symlink
1121 stack in ``walk_component()`` immediately whe
1122 ``walk_component()`` is also the last piece o
1123 old symlink as it walks that last component.
1124 convenient for ``walk_component()`` to releas
1125 the references just before pushing the refere
1126 new symlink. It is guided in this by three f
1127 forbids it from following a symlink if it fin
1128 which indicates that it is yet too early to r
1129 current symlink, and ``WALK_TRAILING`` which
1130 component of the lookup, so we will check use
1131 decide whether follow it when it is a symlink
1132 check if we have privilege to follow it.
1133
1134 Symlinks with no final component
1135 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1136
1137 A pair of special-case symlinks deserve a lit
1138 Both result in a new ``struct path`` (with mo
1139 up in the ``nameidata``, and result in pick_l
1140
1141 The more obvious case is a symlink to "``/``"
1142 with "``/``" are detected in pick_link() whic
1143 to point to the effective filesystem root. I
1144 contains "``/``" then there is nothing more t
1145 so ``NULL`` is returned to indicate that the
1146 the stack frame discarded.
1147
1148 The other case involves things in ``/proc`` t
1149 aren't really (and are therefore commonly ref
1150
1151 $ ls -l /proc/self/fd/1
1152 lrwx------ 1 neilb neilb 64 Jun 13 10:19
1153
1154 Every open file descriptor in any process is
1155 something that looks like a symlink. It is r
1156 target file, not just the name of it. When y
1157 objects you get a name that might refer to th
1158 has been unlinked or mounted over. When ``wa
1159 one of these, the ``->get_link()`` method in
1160 a string name, but instead calls nd_jump_link
1161 ``nameidata`` in place to point to that targe
1162 returns ``NULL``. Again there is no final co
1163 returns ``NULL``.
1164
1165 Following the symlink in the final component
1166 --------------------------------------------
1167
1168 All this leads to ``link_path_walk()`` walkin
1169 following all symbolic links it finds, until
1170 component. This is just returned in the ``la
1171 For some callers, this is all they need; they
1172 ``last`` name if it doesn't exist or give an
1173 callers will want to follow a symlink if one
1174 apply special handling to the last component
1175 than just the last component of the original
1176 potentially need to call ``link_path_walk()``
1177 successive symlinks until one is found that d
1178 symlink.
1179
1180 This case is handled by relevant callers of l
1181 path_lookupat(), path_openat() using a loop t
1182 and then handles the final component by calli
1183 lookup_last(). If it is a symlink that needs
1184 open_last_lookups() or lookup_last() will set
1185 return the path so that the loop repeats, cal
1186 link_path_walk() again. This could loop as m
1187 component of each symlink is another symlink.
1188
1189 Of the various functions that examine the fin
1190 open_last_lookups() is the most interesting a
1191 with do_open() for opening a file. Part of o
1192 with ``i_rwsem`` held and this part is in a s
1193
1194 Explaining open_last_lookups() and do_open()
1195 of this article, but a few highlights should
1196 the code.
1197
1198 1. Rather than just finding the target file,
1199 open_last_lookup() to open
1200 it. If the file was found in the dcache,
1201 this. If not, then ``lookup_open()`` will
1202 the filesystem provides it) to combine the
1203 will perform the separate ``i_op->lookup()
1204 directly. In the later case the actual "o
1205 created file will be performed by vfs_open
1206 were found in the dcache.
1207
1208 2. vfs_open() can fail with ``-EOPENSTALE`` i
1209 wasn't quite current enough. If it's in R
1210 otherwise ``-ESTALE`` is returned. When `
1211 retry with ``LOOKUP_REVAL`` flag set.
1212
1213 3. An open with O_CREAT **does** follow a sym
1214 unlike other creation system calls (like `
1215
1216 ln -s bar /tmp/foo
1217 echo hello > /tmp/foo
1218
1219 will create a file called ``/tmp/bar``. T
1220 ``O_EXCL`` is set but otherwise is handled
1221 like for a non-creating open: lookup_last(
1222 returns a non ``NULL`` value, and link_pat
1223 open process continues on the symlink that
1224
1225 Updating the access time
1226 ------------------------
1227
1228 We previously said of RCU-walk that it would
1229 no counts, leave no footprints." We have sin
1230 "footprints" can be needed when handling syml
1231 reference (or even a memory allocation) may b
1232 footprints are best kept to a minimum.
1233
1234 One other place where walking down a symlink
1235 footprints in a way that doesn't affect direc
1236 In Unix (and Linux) every filesystem object h
1237 time", or "``atime``". Passing through a dir
1238 within is not considered to be an access for
1239 ``atime``; only listing the contents of a dir
1240 Symlinks are different it seems. Both readin
1241 and looking up a symlink on the way to some o
1242 update the atime on that symlink.
1243
1244 .. _clearest statement: https://pubs.opengrou
1245
1246 It is not clear why this is the case; POSIX h
1247 subject. The `clearest statement`_ is that,
1248 updates a timestamp in a place not specified
1249 documented "except that any changes caused by
1250 not be documented". This seems to imply that
1251 care about access-time updates during pathnam
1252
1253 .. _Linux 1.3.87: https://git.kernel.org/cgit
1254
1255 An examination of history shows that prior to
1256 filesystem, at least, didn't update atime whe
1257 Unfortunately we have no record of why that b
1258
1259 In any case, access time must now be updated
1260 quite complex. Trying to stay in RCU-walk wh
1261 avoided. Fortunately it is often permitted t
1262 update. Because ``atime`` updates cause perf
1263 areas, Linux supports the ``relatime`` mount
1264 limits the updates of ``atime`` to once per d
1265 being changed (and symlinks never change once
1266 ``relatime``, many filesystems record ``atime
1267 granularity, so only one update per second is
1268
1269 It is easy to test if an ``atime`` update is
1270 mode and, if it isn't, the update can be skip
1271 continues. Only when an ``atime`` update is
1272 path walk drop down to REF-walk. All of this
1273 ``get_link()`` function.
1274
1275 A few flags
1276 -----------
1277
1278 A suitable way to wrap up this tour of pathna
1279 the various flags that can be stored in the `
1280 lookup process. Many of these are only meani
1281 component, others reflect the current state o
1282 apply restrictions to all path components enc
1283
1284 And then there is ``LOOKUP_EMPTY``, which doe
1285 the others. If this is not set, an empty pat
1286 very early on. If it is set, empty pathnames
1287 an error.
1288
1289 Global state flags
1290 ~~~~~~~~~~~~~~~~~~
1291
1292 We have already met two global state flags: `
1293 ``LOOKUP_REVAL``. These select between one o
1294 to lookup: RCU-walk, REF-walk, and REF-walk w
1295
1296 ``LOOKUP_PARENT`` indicates that the final co
1297 yet. This is primarily used to tell the audi
1298 context of a particular access being audited.
1299
1300 ``ND_ROOT_PRESET`` indicates that the ``root`
1301 provided by the caller, so it shouldn't be re
1302 longer needed.
1303
1304 ``ND_JUMPED`` means that the current dentry w
1305 it had the right name but for some other reas
1306 following "``..``", following a symlink to ``
1307 or accessing a "``/proc/$PID/fd/$FD``" symlin
1308 link"). In this case the filesystem has not b
1309 name (with ``d_revalidate()``). In such case
1310 to be revalidated, so ``d_op->d_weak_revalida
1311 ``ND_JUMPED`` is set when the look completes
1312 final component or, when creating, unlinking,
1313
1314 Resolution-restriction flags
1315 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1316
1317 In order to allow userspace to protect itself
1318 and attack scenarios involving changing path
1319 available which apply restrictions to all pat
1320 path lookup. These flags are exposed through
1321
1322 ``LOOKUP_NO_SYMLINKS`` blocks all symlink tra
1323 This is distinctly different from ``LOOKUP_FO
1324 relates to restricting the following of trail
1325
1326 ``LOOKUP_NO_MAGICLINKS`` blocks all magic-lin
1327 ensure that they return errors from ``nd_jump
1328 ``LOOKUP_NO_MAGICLINKS`` and other magic-link
1329
1330 ``LOOKUP_NO_XDEV`` blocks all ``vfsmount`` tr
1331 bind-mounts and ordinary mounts). Note that t
1332 lookup is determined by the first mountpoint
1333 absolute paths start with the ``vfsmount`` of
1334 with the ``dfd``'s ``vfsmount``. Magic-links
1335 ``vfsmount`` of the path is unchanged.
1336
1337 ``LOOKUP_BENEATH`` blocks any path components
1338 starting point of the resolution. This is don
1339 as well as blocking ".." if it would jump out
1340 ``rename_lock`` and ``mount_lock`` are used t
1341 resolution of "..". Magic-links are also bloc
1342
1343 ``LOOKUP_IN_ROOT`` resolves all path componen
1344 were the filesystem root. ``nd_jump_root()``
1345 the starting point, and ".." at the starting
1346 ``LOOKUP_BENEATH``, ``rename_lock`` and ``mou
1347 attacks against ".." resolution. Magic-links
1348
1349 Final-component flags
1350 ~~~~~~~~~~~~~~~~~~~~~
1351
1352 Some of these flags are only set when the fin
1353 considered. Others are only checked for when
1354 component.
1355
1356 ``LOOKUP_AUTOMOUNT`` ensures that, if the fin
1357 point, then the mount is triggered. Some ope
1358 anyway, but operations like ``stat()`` delibe
1359 needs to trigger the mount but otherwise beha
1360 it sets ``LOOKUP_AUTOMOUNT``, as does "``quot
1361 "``mount --bind``".
1362
1363 ``LOOKUP_FOLLOW`` has a similar function to `
1364 symlinks. Some system calls set or clear it
1365 others have API flags such as ``AT_SYMLINK_FO
1366 ``UMOUNT_NOFOLLOW`` to control it. Its effec
1367 ``WALK_GET`` that we already met, but it is u
1368
1369 ``LOOKUP_DIRECTORY`` insists that the final c
1370 Various callers set this and it is also set w
1371 is found to be followed by a slash.
1372
1373 Finally ``LOOKUP_OPEN``, ``LOOKUP_CREATE``, `
1374 ``LOOKUP_RENAME_TARGET`` are not used directl
1375 available to the filesystem and particularly
1376 method. A filesystem can choose not to bothe
1377 if it knows that it will be asked to open or
1378 These flags were previously useful for ``->lo
1379 introduction of ``->atomic_open()`` they are
1380
1381 End of the road
1382 ---------------
1383
1384 Despite its complexity, all this pathname loo
1385 in good shape - various parts are certainly e
1386 than even a couple of releases ago. But that
1387 "finished". As already mentioned, RCU-walk
1388 symlinks that are stored in the inode so, whi
1389 symlinks, it doesn't help with NFS, XFS, or B
1390 is not likely to be long delayed.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.