2 * Copyright (c) 2002-2014 Apple Inc. All rights reserved.
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
31 #include <sys/appleapiopts.h>
34 #ifdef __APPLE_API_PRIVATE
36 #include <sys/types.h>
37 #include <sys/queue.h>
39 #include <sys/vnode.h>
40 #include <sys/quota.h>
42 #include <kern/locks.h>
44 #include <hfs/hfs_catalog.h>
45 #include <hfs/rangelist.h>
47 #include <sys/decmpfs.h>
50 #include <sys/cprotect.h>
55 * The filefork is used to represent an HFS file fork (data or resource).
56 * Reading or writing any of these fields requires holding cnode lock.
59 struct cnode
*ff_cp
; /* cnode associated with this fork */
60 struct rl_head ff_invalidranges
; /* Areas of disk that should read back as zeroes */
62 void *ffu_sysfileinfo
; /* additional info for system files */
63 char *ffu_symlinkptr
; /* symbolic link pathname */
65 struct cat_fork ff_data
; /* fork data (size, extents) */
67 typedef struct filefork filefork_t
;
70 #define HFS_TEMPLOOKUP_NAMELEN 32
73 * Catalog Lookup struct (runtime)
75 * This is used so that when we need to malloc a container for a catalog
76 * lookup operation, we can acquire memory for everything in one fell swoop
77 * as opposed to putting many of these objects on the stack. The cat_fork
78 * data structure can take up 100+bytes easily, and that can add to stack
81 * As a result, we use this to easily pass around the memory needed for a
84 struct cat_lookup_buffer
{
85 struct cat_desc lookup_desc
;
86 struct cat_attr lookup_attr
;
87 struct filefork lookup_fork
;
88 struct componentname lookup_cn
;
89 char lookup_name
[HFS_TEMPLOOKUP_NAMELEN
]; /* for open-unlinked paths only */
93 /* Aliases for common fields */
94 #define ff_size ff_data.cf_size
95 #define ff_new_size ff_data.cf_new_size
96 #define ff_clumpsize ff_data.cf_clump
97 #define ff_bytesread ff_data.cf_bytesread
98 #define ff_blocks ff_data.cf_blocks
99 #define ff_extents ff_data.cf_extents
100 #define ff_unallocblocks ff_data.cf_vblocks
102 #define ff_symlinkptr ff_union.ffu_symlinkptr
103 #define ff_sysfileinfo ff_union.ffu_sysfileinfo
106 /* The btree code still needs these... */
107 #define fcbEOF ff_size
108 #define fcbExtents ff_extents
109 #define fcbBTCBPtr ff_sysfileinfo
111 typedef u_int8_t atomicflag_t
;
115 * Hardlink Origin (for hardlinked directories).
118 TAILQ_ENTRY(linkorigin
) lo_link
; /* chain */
119 void * lo_thread
; /* thread that performed the lookup */
120 cnid_t lo_cnid
; /* hardlink's cnid */
121 cnid_t lo_parentcnid
; /* hardlink's parent cnid */
123 typedef struct linkorigin linkorigin_t
;
125 #define MAX_CACHED_ORIGINS 10
126 #define MAX_CACHED_FILE_ORIGINS 8
129 * The cnode is used to represent each active (or recently active)
130 * file or directory in the HFS filesystem.
132 * Reading or writing any of these fields requires holding c_lock.
135 lck_rw_t c_rwlock
; /* cnode's lock */
136 thread_t c_lockowner
; /* cnode's lock owner (exclusive case only) */
137 lck_rw_t c_truncatelock
; /* protects file from truncation during read/write */
138 thread_t c_truncatelockowner
; /* truncate lock owner (exclusive case only) */
139 LIST_ENTRY(cnode
) c_hash
; /* cnode's hash chain */
140 u_int32_t c_flag
; /* cnode's runtime flags */
141 u_int32_t c_hflag
; /* cnode's flags for maintaining hash - protected by global hash lock */
142 struct vnode
*c_vp
; /* vnode for data fork or dir */
143 struct vnode
*c_rsrc_vp
; /* vnode for resource fork */
144 struct dquot
*c_dquot
[MAXQUOTAS
]; /* cnode's quota info */
145 u_int32_t c_childhint
; /* catalog hint for children (small dirs only) */
146 u_int32_t c_dirthreadhint
; /* catalog hint for directory's thread rec */
147 struct cat_desc c_desc
; /* cnode's descriptor */
148 struct cat_attr c_attr
; /* cnode's attributes */
149 TAILQ_HEAD(hfs_originhead
, linkorigin
) c_originlist
; /* hardlink origin cache */
150 TAILQ_HEAD(hfs_hinthead
, directoryhint
) c_hintlist
; /* readdir directory hint list */
151 int16_t c_dirhinttag
; /* directory hint tag */
153 int16_t cu_dirhintcnt
; /* directory hint count */
154 int16_t cu_syslockcount
; /* system file use only */
156 u_int32_t c_dirchangecnt
; /* changes each insert/delete (in-core only) */
157 struct filefork
*c_datafork
; /* cnode's data fork */
158 struct filefork
*c_rsrcfork
; /* cnode's rsrc fork */
159 atomicflag_t c_touch_acctime
;
160 atomicflag_t c_touch_chgtime
;
161 atomicflag_t c_touch_modtime
;
163 // The following flags are protected by the truncate lock
166 bool c_need_dvnode_put_after_truncate_unlock
: 1;
167 bool c_need_rvnode_put_after_truncate_unlock
: 1;
169 bool c_need_decmpfs_reset
: 1;
176 decmpfs_cnode
*c_decmp
;
177 #endif /* HFS_COMPRESSION */
179 cprotect_t c_cpentry
; /* content protection data */
183 typedef struct cnode cnode_t
;
185 /* Aliases for common cnode fields */
186 #define c_cnid c_desc.cd_cnid
187 #define c_hint c_desc.cd_hint
188 #define c_parentcnid c_desc.cd_parentcnid
189 #define c_encoding c_desc.cd_encoding
191 #define c_fileid c_attr.ca_fileid
192 #define c_mode c_attr.ca_mode
193 #define c_linkcount c_attr.ca_linkcount
194 #define c_uid c_attr.ca_uid
195 #define c_gid c_attr.ca_gid
196 #define c_rdev c_attr.ca_union1.cau_rdev
197 #define c_atime c_attr.ca_atime
198 #define c_mtime c_attr.ca_mtime
199 #define c_ctime c_attr.ca_ctime
200 #define c_itime c_attr.ca_itime
201 #define c_btime c_attr.ca_btime
202 #define c_bsdflags c_attr.ca_flags
203 #define c_finderinfo c_attr.ca_finderinfo
204 #define c_blocks c_attr.ca_union2.cau_blocks
205 #define c_entries c_attr.ca_union2.cau_entries
206 #define c_zftimeout c_childhint
208 #define c_dirhintcnt c_union.cu_dirhintcnt
209 #define c_syslockcount c_union.cu_syslockcount
212 /* hash maintenance flags kept in c_hflag and protected by hfs_chash_mutex */
213 #define H_ALLOC 0x00001 /* CNode is being allocated */
214 #define H_ATTACH 0x00002 /* CNode is being attached to by another vnode */
215 #define H_TRANSIT 0x00004 /* CNode is getting recycled */
216 #define H_WAITING 0x00008 /* CNode is being waited for */
220 * Runtime cnode flags (kept in c_flag)
222 #define C_NEED_RVNODE_PUT 0x0000001 /* Need to do a vnode_put on c_rsrc_vp after the unlock */
223 #define C_NEED_DVNODE_PUT 0x0000002 /* Need to do a vnode_put on c_vp after the unlock */
224 #define C_ZFWANTSYNC 0x0000004 /* fsync requested and file has holes */
225 #define C_FROMSYNC 0x0000008 /* fsync was called from sync */
227 #define C_MODIFIED 0x0000010 /* CNode has been modified */
228 #define C_NOEXISTS 0x0000020 /* CNode has been deleted, catalog entry is gone */
229 #define C_DELETED 0x0000040 /* CNode has been marked to be deleted */
230 #define C_HARDLINK 0x0000080 /* CNode is a hard link (file or dir) */
232 #define C_FORCEUPDATE 0x0000100 /* force the catalog entry update */
233 #define C_HASXATTRS 0x0000200 /* cnode has extended attributes */
234 #define C_NEG_ENTRIES 0x0000400 /* directory has negative name entries */
236 * For C_SSD_STATIC: SSDs may want to deal with the file payload data in a
237 * different manner knowing that the content is not likely to be modified. This is
238 * purely advisory at the HFS level, and is not maintained after the cnode goes out of core.
240 #define C_SSD_STATIC 0x0000800 /* Assume future writes contain static content */
242 #define C_NEED_DATA_SETSIZE 0x0001000 /* Do a ubc_setsize(0) on c_rsrc_vp after the unlock */
243 #define C_NEED_RSRC_SETSIZE 0x0002000 /* Do a ubc_setsize(0) on c_vp after the unlock */
244 #define C_DIR_MODIFICATION 0x0004000 /* Directory is being modified, wait for lookups */
245 #define C_ALWAYS_ZEROFILL 0x0008000 /* Always zero-fill the file on an fsync */
247 #define C_RENAMED 0x0010000 /* cnode was deleted as part of rename; C_DELETED should also be set */
248 #define C_NEEDS_DATEADDED 0x0020000 /* cnode needs date-added written to the finderinfo bit */
249 #define C_BACKINGSTORE 0x0040000 /* cnode is a backing store for an existing or currently-mounting filesystem */
252 * This flag indicates the cnode might be dirty because it
253 * was mapped writable so if we get any page-outs, update
254 * the modification and change times.
256 #define C_MIGHT_BE_DIRTY_FROM_MAPPING 0x0080000
259 * For C_SSD_GREEDY_MODE: SSDs may want to write the file payload data using the greedy mode knowing
260 * that the content needs to be written out to the disk quicker than normal at the expense of storage efficiency.
261 * This is purely advisory at the HFS level, and is not maintained after the cnode goes out of core.
263 #define C_SSD_GREEDY_MODE 0x0100000 /* Assume future writes are recommended to be written in SLC mode */
265 /* 0x0200000 is currently unused */
267 #define C_IO_ISOCHRONOUS 0x0400000 /* device-specific isochronous throughput I/O */
269 #define ZFTIMELIMIT (5 * 60)
272 * The following is the "invisible" bit from the fdFlags field
273 * in the FndrFileInfo.
275 enum { kFinderInvisibleMask
= 1 << 14 };
279 * Convert between cnode pointers and vnode pointers
281 #define VTOC(vp) ((struct cnode *)vnode_fsnode((vp)))
283 #define CTOV(cp,rsrc) (((rsrc) && S_ISREG((cp)->c_mode)) ? \
284 (cp)->c_rsrc_vp : (cp)->c_vp)
287 * Convert between vnode pointers and file forks
289 * Note: no CTOF since that is ambiguous
292 #define FTOC(fp) ((fp)->ff_cp)
294 #define VTOF(vp) ((vp) == VTOC((vp))->c_rsrc_vp ? \
295 VTOC((vp))->c_rsrcfork : \
296 VTOC((vp))->c_datafork)
298 #define VCTOF(vp, cp) ((vp) == (cp)->c_rsrc_vp ? \
302 #define FTOV(fp) ((fp) == FTOC(fp)->c_rsrcfork ? \
303 FTOC(fp)->c_rsrc_vp : \
307 * This is a helper function used for determining whether or not a cnode has become open
308 * unlinked in between the time we acquired its vnode and the time we acquire the cnode lock
309 * to start manipulating it. Due to the SMP nature of VFS, it is probably necessary to
310 * use this macro every time we acquire a cnode lock, as the content of the Cnode may have
311 * been modified in betweeen the lookup and a VNOP. Whether or not to call this is dependent
312 * upon the VNOP in question. Sometimes it is OK to use an open-unlinked file, for example, in,
313 * reading. But other times, such as on the source of a VNOP_RENAME, it should be disallowed.
315 int hfs_checkdeleted(struct cnode
*cp
);
318 * Test for a resource fork
320 #define FORK_IS_RSRC(fp) ((fp) == FTOC(fp)->c_rsrcfork)
322 #define VNODE_IS_RSRC(vp) ((vp) == VTOC((vp))->c_rsrc_vp)
326 * VTOCMP(vp) returns a pointer to vp's decmpfs_cnode; this could be NULL
327 * if the file is not compressed or if hfs_file_is_compressed() hasn't
328 * yet been called on this file.
330 #define VTOCMP(vp) (VTOC((vp))->c_decmp)
331 int hfs_file_is_compressed(struct cnode
*cp
, int skiplock
);
332 int hfs_uncompressed_size_of_compressed_file(struct hfsmount
*hfsmp
, struct vnode
*vp
, cnid_t fid
, off_t
*size
, int skiplock
);
333 int hfs_hides_rsrc(vfs_context_t ctx
, struct cnode
*cp
, int skiplock
);
334 int hfs_hides_xattr(vfs_context_t ctx
, struct cnode
*cp
, const char *name
, int skiplock
);
337 #define ATIME_ONDISK_ACCURACY 300
340 /* This overlays the FileID portion of NFS file handles. */
342 u_int32_t hfsfid_cnid
; /* Catalog node ID. */
343 u_int32_t hfsfid_gen
; /* Generation number (create date). */
347 /* Get new default vnode */
348 extern int hfs_getnewvnode(struct hfsmount
*hfsmp
, struct vnode
*dvp
, struct componentname
*cnp
,
349 struct cat_desc
*descp
, int flags
, struct cat_attr
*attrp
,
350 struct cat_fork
*forkp
, struct vnode
**vpp
, int *out_flags
);
352 /* Input flags for hfs_getnewvnode */
354 #define GNV_WANTRSRC 0x01 /* Request the resource fork vnode. */
355 #define GNV_SKIPLOCK 0x02 /* Skip taking the cnode lock (when getting resource fork). */
356 #define GNV_CREATE 0x04 /* The vnode is for a newly created item. */
357 #define GNV_NOCACHE 0x08 /* Delay entering this item in the name cache */
359 /* Output flags for hfs_getnewvnode */
360 #define GNV_CHASH_RENAMED 0x01 /* The cnode was renamed in-flight */
361 #define GNV_CAT_DELETED 0x02 /* The cnode was deleted from the catalog */
362 #define GNV_NEW_CNODE 0x04 /* We are vending out a newly initialized cnode */
363 #define GNV_CAT_ATTRCHANGED 0x08 /* Something in struct cat_attr changed in between cat_lookups */
365 /* Touch cnode times based on c_touch_xxx flags */
366 extern void hfs_touchtimes(struct hfsmount
*, struct cnode
*);
367 extern void hfs_write_dateadded (struct cat_attr
*cattrp
, u_int32_t dateadded
);
368 extern u_int32_t
hfs_get_dateadded (struct cnode
*cp
);
369 extern u_int32_t
hfs_get_dateadded_from_blob(const uint8_t * /* finderinfo */, mode_t
/* mode */);
371 /* Gen counter methods */
372 extern void hfs_write_gencount(struct cat_attr
*cattrp
, uint32_t gencount
);
373 extern uint32_t hfs_get_gencount(struct cnode
*cp
);
374 extern uint32_t hfs_incr_gencount (struct cnode
*cp
);
375 extern uint32_t hfs_get_gencount_from_blob(const uint8_t * /* finderinfo */, mode_t
/* mode */);
377 /* Document id methods */
378 extern uint32_t hfs_get_document_id(struct cnode
* /* cp */);
379 extern uint32_t hfs_get_document_id_from_blob(const uint8_t * /* finderinfo */, mode_t
/* mode */);
381 /* Zero-fill file and push regions out to disk */
383 // Use this flag if you're going to sync later
384 HFS_FILE_DONE_NO_SYNC
= 1,
386 typedef uint32_t hfs_file_done_opts_t
;
387 extern int hfs_filedone(struct vnode
*vp
, vfs_context_t context
,
388 hfs_file_done_opts_t opts
);
391 * HFS cnode hash functions.
393 extern void hfs_chashinit(void);
394 extern void hfs_chashinit_finish(struct hfsmount
*hfsmp
);
395 extern void hfs_delete_chash(struct hfsmount
*hfsmp
);
396 extern int hfs_chashremove(struct hfsmount
*hfsmp
, struct cnode
*cp
);
397 extern void hfs_chash_abort(struct hfsmount
*hfsmp
, struct cnode
*cp
);
398 extern void hfs_chash_rehash(struct hfsmount
*hfsmp
, struct cnode
*cp1
, struct cnode
*cp2
);
399 extern void hfs_chashwakeup(struct hfsmount
*hfsmp
, struct cnode
*cp
, int flags
);
400 extern void hfs_chash_mark_in_transit(struct hfsmount
*hfsmp
, struct cnode
*cp
);
402 extern struct vnode
* hfs_chash_getvnode(struct hfsmount
*hfsmp
, ino_t inum
, int wantrsrc
,
403 int skiplock
, int allow_deleted
);
404 extern struct cnode
* hfs_chash_getcnode(struct hfsmount
*hfsmp
, ino_t inum
, struct vnode
**vpp
,
405 int wantrsrc
, int skiplock
, int *out_flags
, int *hflags
);
406 extern int hfs_chash_snoop(struct hfsmount
*, ino_t
, int, int (*)(const cnode_t
*, void *), void *);
407 extern int hfs_valid_cnode(struct hfsmount
*hfsmp
, struct vnode
*dvp
, struct componentname
*cnp
,
408 cnid_t cnid
, struct cat_attr
*cattr
, int *error
);
410 extern int hfs_chash_set_childlinkbit(struct hfsmount
*hfsmp
, cnid_t cnid
);
413 * HFS cnode lock functions.
417 * 1. cnode truncate lock (if needed) -- see below for more on this
419 * + hfs_vnop_pagein/out handles recursive use of this lock (by
420 * using flag option HFS_LOCK_SKIP_IF_EXCLUSIVE) although there
421 * are issues with this (see #16620278).
423 * + If locking multiple cnodes then the truncate lock must be taken on
424 * both (in address order), before taking the cnode locks.
426 * 2. cnode lock (in parent-child order if related, otherwise by address order)
428 * 3. journal (if needed)
430 * 4. system files (as needed)
432 * A. Catalog B-tree file
433 * B. Attributes B-tree file
434 * C. Startup file (if there is one)
435 * D. Allocation Bitmap file (always exclusive, supports recursion)
436 * E. Overflow Extents B-tree file (always exclusive, supports recursion)
438 * 5. hfs mount point (always last)
441 * I. HFS cnode hash lock (must not acquire any new locks while holding this lock, always taken last)
445 * -- The Truncate Lock --
447 * The truncate lock is used for a few purposes (more than its name
448 * might suggest). The first thing to note is that the cnode lock
449 * cannot be held whilst issuing any I/O other than metadata changes,
450 * so the truncate lock, in either shared or exclusive form, must
451 * usually be held in these cases. This includes calls to ubc_setsize
452 * where the new size is less than the current size known to the VM
453 * subsystem (for two reasons: a) because reaping pages can block
454 * (e.g. on pages that are busy or being cleaned); b) reaping pages
455 * might require page-in for tasks that have that region mapped
456 * privately). The same applies to other calls into the VM subsystem.
458 * Here are some (but not necessarily all) cases that the truncate
461 * + When reading and writing a file, we hold the truncate lock
462 * shared to ensure that the underlying blocks cannot be deleted
463 * and on systems that use content protection, this also ensures
464 * the keys remain valid (which might be being used by the
465 * underlying layers).
467 * + We need to protect against the following sequence of events:
469 * A file is initially size X. A thread issues an append to that
470 * file. Another thread truncates the file and then extends it
471 * to a a new size Y. Now the append can be applied at offset X
472 * and then the data is lost when the file is truncated; or it
473 * could be applied after the truncate, i.e. at offset 0; or it
474 * can be applied at offset Y. What we *cannot* do is apply the
475 * append at offset X and for the data to be visible at the end.
476 * (Note that we are free to choose when we apply the append
479 * To solve this, we keep things simple and take the truncate lock
480 * exclusively in order to sequence the append with other size
481 * changes. Therefore any size change must take the truncate lock
484 * (N.B. we could do better and allow readers to run concurrently
485 * during the append and other size changes.)
487 * So here are the rules:
489 * + If you plan to change ff_size, you must take the truncate lock
490 * exclusively, *but* be careful what I/O you do whilst you have
491 * the truncate lock exclusively and try and avoid it if you can:
492 * if the VM subsystem tries to do something with some pages on a
493 * different thread and you try and do some I/O with those same
494 * pages, we will deadlock. (See #16620278.)
496 * + If you do anything that requires blocks to not be deleted or
497 * encrpytion keys to remain valid, you must take the truncate lock
500 * + And it follows therefore, that if you want to delete blocks or
501 * delete keys, you must take the truncate lock exclusively.
503 * N.B. ff_size is actually protected by the cnode lock and so you
504 * must hold the cnode lock exclusively to change it and shared to
511 HFS_EXCLUSIVE_LOCK
= 2
514 /* Option flags for cnode and truncate lock functions */
516 HFS_LOCK_DEFAULT
= 0x0, /* Default flag, no options provided */
517 HFS_LOCK_ALLOW_NOEXISTS
= 0x1, /* Allow locking of all cnodes, including cnode marked deleted with no catalog entry */
518 HFS_LOCK_SKIP_IF_EXCLUSIVE
= 0x2, /* Skip locking if the current thread already holds the lock exclusive */
520 // Used when you do not want to check return from hfs_lock
521 HFS_LOCK_ALWAYS
= HFS_LOCK_ALLOW_NOEXISTS
,
523 #define HFS_SHARED_OWNER (void *)0xffffffff
525 void hfs_lock_always(cnode_t
*cnode
, enum hfs_locktype
);
526 int hfs_lock(struct cnode
*, enum hfs_locktype
, enum hfs_lockflags
);
527 int hfs_lockpair(struct cnode
*, struct cnode
*, enum hfs_locktype
);
528 int hfs_lockfour(struct cnode
*, struct cnode
*, struct cnode
*, struct cnode
*,
529 enum hfs_locktype
, struct cnode
**);
531 void hfs_unlock(struct cnode
*);
532 void hfs_unlockpair(struct cnode
*, struct cnode
*);
533 void hfs_unlockfour(struct cnode
*, struct cnode
*, struct cnode
*, struct cnode
*);
535 void hfs_lock_truncate(struct cnode
*, enum hfs_locktype
, enum hfs_lockflags
);
536 void hfs_unlock_truncate(struct cnode
*, enum hfs_lockflags
);
537 int hfs_try_trunclock(struct cnode
*, enum hfs_locktype
, enum hfs_lockflags
);
539 void hfs_clear_might_be_dirty_flag(cnode_t
*cp
);
541 // cnode must be locked
542 static inline __attribute__((pure
))
543 bool hfs_has_rsrc(const cnode_t
*cp
)
546 return cp
->c_rsrcfork
->ff_blocks
> 0;
548 return cp
->c_datafork
&& cp
->c_blocks
> cp
->c_datafork
->ff_blocks
;
551 #endif /* __APPLE_API_PRIVATE */
554 #endif /* ! _HFS_CNODE_H_ */