xref: /titanic_41/usr/src/uts/common/sys/fs/ufs_inode.h (revision 672986541be54a7a471bb088e60780c37e371d7e)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	  All Rights Reserved  	*/
28 
29 /*
30  * University Copyright- Copyright (c) 1982, 1986, 1988
31  * The Regents of the University of California
32  * All Rights Reserved
33  *
34  * University Acknowledgment- Portions of this document are derived from
35  * software developed by the University of California, Berkeley, and its
36  * contributors.
37  */
38 
39 #ifndef	_SYS_FS_UFS_INODE_H
40 #define	_SYS_FS_UFS_INODE_H
41 
42 #pragma ident	"%Z%%M%	%I%	%E% SMI"
43 
44 #include <sys/isa_defs.h>
45 #include <sys/fbuf.h>
46 #include <sys/fdbuffer.h>
47 #include <sys/fcntl.h>
48 #include <sys/uio.h>
49 #include <sys/t_lock.h>
50 #include <sys/thread.h>
51 #include <sys/cred.h>
52 #include <sys/time.h>
53 #include <sys/types32.h>
54 #include <sys/fs/ufs_fs.h>
55 #include <sys/fs/ufs_lockfs.h>
56 #include <sys/fs/ufs_trans.h>
57 #include <sys/kstat.h>
58 #include <sys/fs/ufs_acl.h>
59 #include <sys/fs/ufs_panic.h>
60 #include <sys/dnlc.h>
61 
62 #ifdef	__cplusplus
63 extern "C" {
64 #endif
65 
66 /*
67  * The I node is the focus of all local file activity in UNIX.
68  * There is a unique inode allocated for each active file,
69  * each current directory, each mounted-on file, each mapping,
70  * and the root.  An inode is `named' by its dev/inumber pair.
71  * Data in icommon is read in from permanent inode on volume.
72  *
73  * Each inode has 5 locks associated with it:
74  *	i_rwlock:	Serializes ufs_write and ufs_setattr request
75  *			and allows ufs_read requests to proceed in parallel.
76  *			Serializes reads/updates to directories.
77  *	vfs_dqrwlock:	Manages quota sub-system quiescence.  See below.
78  *	i_contents:	Protects almost all of the fields in the inode
79  *			except for those listed below. When held
80  *			in writer mode also protects those fields
81  *			listed under i_tlock.
82  *	i_tlock:	When i_tlock is held with the i_contents reader
83  *			lock the i_atime, i_mtime, i_ctime,
84  *			i_delayoff, i_delaylen, i_nextrio, i_writes, i_flag
85  *			i_seq, i_writer & i_mapcnt fields are protected.
86  *			For more i_flag locking info see below.
87  *	ih_lock:	Protects inode hash chain buckets
88  *	ifree_lock:	Protects inode freelist
89  *
90  * Lock ordering:
91  *	i_rwlock > i_contents > i_tlock
92  *	i_rwlock > vfs_dqrwlock > i_contents(writer) > i_tlock
93  *	i_contents > i_tlock
94  *	vfs_dqrwlock > i_contents(writer) > i_tlock
95  *	ih_lock > i_contents > i_tlock
96  *
97  * Making major changes to quota sub-system state, while the file
98  * system is mounted required the addition of another lock.  The
99  * primary lock in the quota sub-system is vfs_dqrwlock in the ufsvfs
100  * structure.  This lock is used to manage quota sub-system quiescence
101  * for a particular file system. Major changes to quota sub-system
102  * state (disabling quotas, enabling quotas, and setting new quota
103  * limits) all require the file system to be quiescent and grabbing
104  * vfs_dqrwlock as writer accomplishes this.  On the other hand,
105  * grabbing vfs_dqrwlock as reader makes the quota sub-system
106  * non-quiescent and lets the quota sub-system know that now is not a
107  * good time to change major quota sub-system state.  Typically
108  * vfs_dqrwlock is grabbed for reading before i_contents is grabbed for
109  * writing.  However, there are cases where vfs_dqrwlock is grabbed for
110  * reading without a corresponding i_contents write grab because there
111  * is no relevant inode.  There are also cases where i_contents is
112  * grabbed for writing when a vfs_dqrwlock read grab is not needed
113  * because the inode changes do not affect quotas.
114  *
115  * Unfortunately, performance considerations have required that we be more
116  * intelligent about using i_tlock when updating i_flag.  Ideally, we would
117  * have simply separated out several of the bits in i_flag into their own
118  * ints to avoid problems.  But, instead, we have implemented the following
119  * rules:
120  *
121  *	o You can update any i_flag field while holding the writer-contents,
122  *	  or by holding the reader-contents AND holding i_tlock.
123  *	  You can only call ITIMES_NOLOCK while holding the writer-contents,
124  *	  or by holding the reader-contents AND holding i_tlock.
125  *
126  *	o For a directory, holding the reader-rw_lock is sufficient for setting
127  *	  IACC.
128  *
129  *	o Races with IREF are avoided by holding the reader contents lock
130  *	  and by holding i_tlock in ufs_rmidle, ufs_putapage, and ufs_getpage.
131  *	  And by holding the writer-contents in ufs_iinactive.
132  *
133  *	o The callers are no longer required to handle the calls to ITIMES
134  *	  and ITIMES_NOLOCK.  The functions that set the i_flag bits are
135  *	  responsible for managing those calls.  The exceptions are the
136  *	  bmap routines.
137  *
138  * SVR4 Extended Fundamental Type (EFT) support:
139  * 	The inode structure has been enhanced to support
140  *	32-bit user-id, 32-bit group-id, and 32-bit device number.
141  *	Standard SVR4 ufs also supports 32-bit mode field.  For the reason
142  *	of backward compatibility with the previous ufs disk format,
143  *	32-bit mode field is not supported.
144  *
145  *	The current inode structure is 100% backward compatible with
146  *	the previous inode structure if no user-id or group-id exceeds
147  *	USHRT_MAX, and no major or minor number of a device number
148  *	stored in an inode exceeds 255.
149  *
150  * Rules for managing i_seq:
151  *	o i_seq is locked under the same rules as i_flag
152  *	o The i_ctime or i_mtime MUST never change without increasing
153  *	  the value of i_seq.
154  *	o You may increase the value of i_seq without the timestamps
155  *	  changing, this may decrease the callers performance but will
156  *	  be functionally correct.
157  *	o The common case is when IUPD or ICHG is set, increase i_seq
158  *	  and immediately call ITIMES* or ufs_iupdat to create a new timestamp.
159  *	o A less common case is the setting of IUPD or ICHG and while still
160  *	  holding the correct lock defer the timestamp and i_seq update
161  *	  until later, but it must still be done before the lock is released.
162  *	  bmap_write is an example of this, where the caller does the update.
163  *	o If multiple changes are being made with the timestamps being
164  *	  updated only at the end, a single increase of i_seq is allowed.
165  *	o If changes are made with IUPD or ICHG being set, but
166  *	  the controlling lock is being dropped before the timestamp is
167  *	  updated, there is a risk that another thread will also change
168  *	  the file, update i_flag, and push just one timestamp update.
169  *	  There is also the risk that another thread calls ITIMES or
170  *	  ufs_iupdat without setting IUPD|ICHG and thus not changing i_seq,
171  *	  this will cause ufs_imark to change the timestamps without changing
172  *	  i_seq. If the controlling lock is dropped, ISEQ must be set to
173  *	  force i_seq to be increased on next ufs_imark, but i_seq MUST still
174  *	  be increased by the original setting thread before its deferred
175  *	  call to ITIMES to insure it is increased the correct number of times.
176  */
177 
178 #define	UID_LONG  (o_uid_t)65535
179 				/* flag value to indicate uid is 32-bit long */
180 #define	GID_LONG  (o_uid_t)65535
181 				/* flag value to indicate gid is 32-bit long */
182 
183 #define	NDADDR	12		/* direct addresses in inode */
184 #define	NIADDR	3		/* indirect addresses in inode */
185 #define	FSL_SIZE (NDADDR + NIADDR - 1) * sizeof (daddr32_t)
186 				/* max fast symbolic name length is 56 */
187 
188 #define	i_fs	i_ufsvfs->vfs_bufp->b_un.b_fs
189 #define	i_vfs	i_vnode->v_vfsp
190 
191 struct 	icommon {
192 	o_mode_t ic_smode;	/*  0: mode and type of file */
193 	short	ic_nlink;	/*  2: number of links to file */
194 	o_uid_t	ic_suid;	/*  4: owner's user id */
195 	o_gid_t	ic_sgid;	/*  6: owner's group id */
196 	u_offset_t ic_lsize;	/*  8: number of bytes in file */
197 #ifdef _KERNEL
198 	struct timeval32 ic_atime;	/* 16: time last accessed */
199 	struct timeval32 ic_mtime;	/* 24: time last modified */
200 	struct timeval32 ic_ctime;	/* 32: last time inode changed */
201 #else
202 	time32_t ic_atime;	/* 16: time last accessed */
203 	int32_t	ic_atspare;
204 	time32_t ic_mtime;	/* 24: time last modified */
205 	int32_t	ic_mtspare;
206 	time32_t ic_ctime;	/* 32: last time inode changed */
207 	int32_t	ic_ctspare;
208 #endif
209 	daddr32_t	ic_db[NDADDR];	/* 40: disk block addresses */
210 	daddr32_t	ic_ib[NIADDR];	/* 88: indirect blocks */
211 	int32_t	ic_flags;	/* 100: cflags */
212 	int32_t	ic_blocks;	/* 104: 512 byte blocks actually held */
213 	int32_t	ic_gen;		/* 108: generation number */
214 	int32_t	ic_shadow;	/* 112: shadow inode */
215 	uid_t	ic_uid;		/* 116: long EFT version of uid */
216 	gid_t	ic_gid;		/* 120: long EFT version of gid */
217 	uint32_t ic_oeftflag;	/* 124: extended attr directory ino, 0 = none */
218 };
219 
220 /*
221  * Large directories can be cached. Directory caching can take the following
222  * states:
223  */
224 typedef enum {
225 	CD_DISABLED_NOMEM = -2,
226 	CD_DISABLED_TOOBIG,
227 	CD_DISABLED,
228 	CD_ENABLED
229 } cachedir_t;
230 
231 /*
232  * Large Files: Note we use the inline functions load_double, store_double
233  * to load and store the long long values of i_size. Therefore the
234  * address of i_size must be eight byte aligned. Kmem_alloc of incore
235  * inode structure makes sure that the structure is 8-byte aligned.
236  * XX64 - reorder this structure?
237  */
238 typedef struct inode {
239 	struct	inode *i_chain[2];	/* must be first */
240 	struct inode *i_freef;	/* free list forward - must be before i_ic */
241 	struct inode *i_freeb;	/* free list back - must be before i_ic */
242 	struct 	icommon	i_ic;	/* Must be here */
243 	struct	vnode *i_vnode;	/* vnode associated with this inode */
244 	struct	vnode *i_devvp;	/* vnode for block I/O */
245 	dev_t	i_dev;		/* device where inode resides */
246 	ino_t	i_number;	/* i number, 1-to-1 with device address */
247 	off_t	i_diroff;	/* offset in dir, where we found last entry */
248 				/* just a hint - no locking needed */
249 	struct ufsvfs *i_ufsvfs; /* incore fs associated with inode */
250 	struct	dquot *i_dquot;	/* quota structure controlling this file */
251 	krwlock_t i_rwlock;	/* serializes write/setattr requests */
252 	krwlock_t i_contents;	/* protects (most of) inode contents */
253 	kmutex_t i_tlock;	/* protects time fields, i_flag */
254 	offset_t i_nextr;	/*					*/
255 				/* next byte read offset (read-ahead)	*/
256 				/*   No lock required			*/
257 				/*					*/
258 	uint_t	i_flag;		/* inode flags */
259 	uint_t	i_seq;		/* modification sequence number */
260 	cachedir_t i_cachedir;	/* Cache this directory on next lookup */
261 				/* - no locking needed  */
262 	long	i_mapcnt;	/* mappings to file pages */
263 	int	*i_map;		/* block list for the corresponding file */
264 	dev_t	i_rdev;		/* INCORE rdev from i_oldrdev by ufs_iget */
265 	size_t	i_delaylen;	/* delayed writes, units=bytes */
266 	offset_t i_delayoff;	/* where we started delaying */
267 	offset_t i_nextrio;	/* where to start the next clust */
268 	long	i_writes;	/* number of outstanding bytes in write q */
269 	kcondvar_t i_wrcv;	/* sleep/wakeup for write throttle */
270 	offset_t i_doff;	/* dinode byte offset in file system */
271 	si_t *i_ufs_acl;	/* pointer to acl entry */
272 	dcanchor_t i_danchor;	/* directory cache anchor */
273 	kthread_t *i_writer;	/* thread which is in window in wrip() */
274 } inode_t;
275 
276 struct dinode {
277 	union {
278 		struct	icommon di_icom;
279 		char	di_size[128];
280 	} di_un;
281 };
282 
283 #define	i_mode		i_ic.ic_smode
284 #define	i_nlink		i_ic.ic_nlink
285 #define	i_uid		i_ic.ic_uid
286 #define	i_gid		i_ic.ic_gid
287 #define	i_smode		i_ic.ic_smode
288 #define	i_suid		i_ic.ic_suid
289 #define	i_sgid		i_ic.ic_sgid
290 
291 #define	i_size		i_ic.ic_lsize
292 #define	i_db		i_ic.ic_db
293 #define	i_ib		i_ic.ic_ib
294 
295 #define	i_atime		i_ic.ic_atime
296 #define	i_mtime		i_ic.ic_mtime
297 #define	i_ctime		i_ic.ic_ctime
298 
299 #define	i_shadow	i_ic.ic_shadow
300 #define	i_oeftflag	i_ic.ic_oeftflag
301 #define	i_blocks	i_ic.ic_blocks
302 #define	i_cflags	i_ic.ic_flags
303 #ifdef _LITTLE_ENDIAN
304 /*
305  * Originally done on x86, but carried on to all other little
306  * architectures, which provides for file system compatibility.
307  */
308 #define	i_ordev		i_ic.ic_db[1]	/* USL SVR4 compatibility */
309 #else
310 #define	i_ordev		i_ic.ic_db[0]	/* was i_oldrdev */
311 #endif
312 #define	i_gen		i_ic.ic_gen
313 #define	i_forw		i_chain[0]
314 #define	i_back		i_chain[1]
315 
316 /* EFT transition aids - obsolete */
317 #define	oEFT_MAGIC	0x90909090
318 #define	di_oeftflag	di_ic.ic_oeftflag
319 
320 #define	di_ic		di_un.di_icom
321 #define	di_mode		di_ic.ic_smode
322 #define	di_nlink	di_ic.ic_nlink
323 #define	di_uid		di_ic.ic_uid
324 #define	di_gid		di_ic.ic_gid
325 #define	di_smode	di_ic.ic_smode
326 #define	di_suid		di_ic.ic_suid
327 #define	di_sgid		di_ic.ic_sgid
328 
329 #define	di_size		di_ic.ic_lsize
330 #define	di_db		di_ic.ic_db
331 #define	di_ib		di_ic.ic_ib
332 
333 #define	di_atime	di_ic.ic_atime
334 #define	di_mtime	di_ic.ic_mtime
335 #define	di_ctime	di_ic.ic_ctime
336 #define	di_cflags	di_ic.ic_flags
337 
338 #ifdef _LITTLE_ENDIAN
339 #define	di_ordev	di_ic.ic_db[1]
340 #else
341 #define	di_ordev	di_ic.ic_db[0]
342 #endif
343 #define	di_shadow	di_ic.ic_shadow
344 #define	di_blocks	di_ic.ic_blocks
345 #define	di_gen		di_ic.ic_gen
346 
347 /* flags */
348 #define	IUPD		0x0001		/* file has been modified */
349 #define	IACC		0x0002		/* inode access time to be updated */
350 #define	IMOD		0x0004		/* inode has been modified */
351 #define	ICHG		0x0008		/* inode has been changed */
352 #define	INOACC		0x0010		/* no access time update in getpage */
353 #define	IMODTIME	0x0020		/* mod time already set */
354 #define	IREF		0x0040		/* inode is being referenced */
355 #define	ISYNC		0x0080		/* do all allocation synchronously */
356 #define	IFASTSYMLNK	0x0100		/* fast symbolic link */
357 #define	IMODACC		0x0200		/* only access time changed; */
358 					/*   filesystem won't become active */
359 #define	IATTCHG		0x0400		/* only size/blocks have changed */
360 #define	IBDWRITE	0x0800		/* the inode has been scheduled for */
361 					/* write operation asynchronously */
362 #define	ISTALE		0x1000		/* inode couldn't be read from disk */
363 #define	IDEL		0x2000		/* inode is being deleted */
364 #define	IDIRECTIO	0x4000		/* attempt directio */
365 #define	ISEQ		0x8000		/* deferred i_seq increase */
366 #define	IJUNKIQ		0x10000		/* on junk idle queue */
367 #define	IQUIET		0x20000		/* No file system full messages */
368 
369 /* cflags */
370 #define	IXATTR		0x0001		/* extended attribute */
371 #define	IFALLOCATE	0x0002		/* fallocate'd file */
372 
373 /* modes */
374 #define	IFMT		0170000		/* type of file */
375 #define	IFIFO		0010000		/* named pipe (fifo) */
376 #define	IFCHR		0020000		/* character special */
377 #define	IFDIR		0040000		/* directory */
378 #define	IFBLK		0060000		/* block special */
379 #define	IFREG		0100000		/* regular */
380 #define	IFLNK		0120000		/* symbolic link */
381 #define	IFSHAD		0130000		/* shadow indode */
382 #define	IFSOCK		0140000		/* socket */
383 #define	IFATTRDIR	0160000		/* Attribute directory */
384 
385 #define	ISUID		04000		/* set user id on execution */
386 #define	ISGID		02000		/* set group id on execution */
387 #define	ISVTX		01000		/* save swapped text even after use */
388 #define	IREAD		0400		/* read, write, execute permissions */
389 #define	IWRITE		0200
390 #define	IEXEC		0100
391 
392 /* specify how the inode info is written in ufs_syncip() */
393 #define	I_SYNC		1		/* wait for the inode written to disk */
394 #define	I_DSYNC		2		/* wait for the inode written to disk */
395 					/* only if IATTCHG is set */
396 #define	I_ASYNC		0		/* don't wait for the inode written */
397 
398 /* flags passed to ufs_itrunc(), indirtrunc(), and free() */
399 #define	I_FREE	0x00000001		/* inode is being freed */
400 #define	I_DIR	0x00000002		/* inode is a directory */
401 #define	I_IBLK	0x00000004		/* indirect block */
402 #define	I_CHEAP	0x00000008		/* cheap free */
403 #define	I_SHAD	0x00000010		/* inode is a shadow inode */
404 #define	I_QUOTA	0x00000020		/* quota file */
405 #define	I_NOCANCEL	0x40		/* Don't cancel these fragments */
406 #define	I_ACCT	0x00000080		/* Update ufsvfs' unreclaimed_blocks */
407 
408 /*
409  * If ufs_dircheckforname() fails to find an entry with the given name,
410  * this "slot" structure holds state for ufs_direnter_*() as to where
411  * there is space to put an entry with that name.
412  * If ufs_dircheckforname() finds an entry with the given name, this structure
413  * holds state for ufs_dirrename() and ufs_dirremove() as to where the
414  * entry is. "status" indicates what ufs_dircheckforname() found:
415  *      NONE            name not found, large enough free slot not found,
416  *      FOUND           name not found, large enough free slot found
417  *      EXIST           name found
418  * If ufs_dircheckforname() fails due to an error, this structure is not
419  * filled in.
420  *
421  * After ufs_dircheckforname() succeeds the values are:
422  *      status  offset          size            fbp, ep
423  *      ------  ------          ----            -------
424  *      NONE    end of dir      needed          not valid
425  *      FOUND   start of entry  of ent          both valid if fbp != NULL
426  *      EXIST   start of entry  of prev ent     valid
427  *
428  * "endoff" is set to 0 if the an entry with the given name is found, or if no
429  * free slot could be found or made; this means that the directory should not
430  * be truncated.  If the entry was found, the search terminates so
431  * ufs_dircheckforname() didn't find out where the last valid entry in the
432  * directory was, so it doesn't know where to cut the directory off; if no free
433  * slot could be found or made, the directory has to be extended to make room
434  * for the new entry, so there's nothing to cut off.
435  * Otherwise, "endoff" is set to the larger of the offset of the last
436  * non-empty entry in the directory, or the offset at which the new entry will
437  * be placed, whichever is larger.  This is used by ufs_diraddentry(); if a new
438  * entry is to be added to the directory, any complete directory blocks at the
439  * end of the directory that contain no non-empty entries are lopped off the
440  * end, thus shrinking the directory dynamically.
441  */
442 typedef enum {NONE, FOUND, EXIST} slotstat_t;
443 struct ufs_slot {
444 	struct	direct *ep;	/* pointer to slot */
445 	struct	fbuf *fbp;	/* dir buf where slot is */
446 	off_t	offset;		/* offset of area with free space */
447 	off_t	endoff;		/* last useful location found in search */
448 	slotstat_t status;	/* status of slot */
449 	int	size;		/* size of area at slotoffset */
450 	int	cached;		/* cached directory */
451 };
452 
453 /*
454  * Statistics on inodes
455  * Not protected by locks
456  */
457 struct instats {
458 	kstat_named_t in_size;		/* current cache size */
459 	kstat_named_t in_maxsize;	/* maximum cache size */
460 	kstat_named_t in_hits;		/* cache hits */
461 	kstat_named_t in_misses;	/* cache misses */
462 	kstat_named_t in_malloc;	/* kmem_alloce'd */
463 	kstat_named_t in_mfree;		/* kmem_free'd */
464 	kstat_named_t in_maxreached;	/* Largest size reached by cache */
465 	kstat_named_t in_frfront;	/* # put at front of freelist */
466 	kstat_named_t in_frback;	/* # put at back of freelist */
467 	kstat_named_t in_qfree;		/* q's to delete thread */
468 	kstat_named_t in_scan;		/* # inodes scanned */
469 	kstat_named_t in_tidles;	/* # inodes idled by idle thread */
470 	kstat_named_t in_lidles;	/* # inodes idled by ufs_lookup */
471 	kstat_named_t in_vidles;	/* # inodes idled by ufs_vget */
472 	kstat_named_t in_kcalloc;	/* # inodes kmem_cache_alloced */
473 	kstat_named_t in_kcfree;	/* # inodes kmem_cache_freed */
474 	kstat_named_t in_poc;		/* # push-on-close's */
475 };
476 
477 #ifdef _KERNEL
478 
479 /*
480  * Extended attributes
481  */
482 
483 #define	XATTR_DIR_NAME	"/@/"
484 extern int	ufs_ninode;		/* high-water mark for inode cache */
485 
486 extern struct vnodeops *ufs_vnodeops;	/* vnode operations for ufs */
487 extern const struct fs_operation_def ufs_vnodeops_template[];
488 
489 /*
490  * Convert between inode pointers and vnode pointers
491  */
492 #define	VTOI(VP)	((struct inode *)(VP)->v_data)
493 #define	ITOV(IP)	((struct vnode *)(IP)->i_vnode)
494 
495 /*
496  * convert to fs
497  */
498 #define	ITOF(IP)	((struct fs *)(IP)->i_fs)
499 
500 /*
501  * Convert between vnode types and inode formats
502  */
503 extern enum vtype	iftovt_tab[];
504 
505 #ifdef notneeded
506 
507 /* Look at sys/mode.h and os/vnode.c */
508 
509 extern int		vttoif_tab[];
510 
511 #endif
512 
513 /*
514  * Mark an inode with the current (unique) timestamp.
515  * (Note that UFS's concept of time only keeps 32 bits of seconds
516  * in the on-disk format).
517  */
518 struct timeval32 iuniqtime;
519 extern kmutex_t ufs_iuniqtime_lock;
520 
521 #define	ITIMES_NOLOCK(ip) ufs_itimes_nolock(ip)
522 
523 #define	ITIMES(ip) { \
524 	mutex_enter(&(ip)->i_tlock); \
525 	ITIMES_NOLOCK(ip); \
526 	mutex_exit(&(ip)->i_tlock); \
527 }
528 
529 /*
530  * The following interfaces are used to do atomic loads and stores
531  * of an inode's i_size, which is a long long data type.
532  *
533  * For LP64, we just to a load or a store - atomicity and alignment
534  * are 8-byte guaranteed.  For x86 there are no such instructions,
535  * so we grab i_contents as reader to get the size; we already hold
536  * it as writer when we're setting the size.
537  */
538 
539 #ifdef _LP64
540 
541 #define	UFS_GET_ISIZE(resultp, ip)	*(resultp) = (ip)->i_size
542 #define	UFS_SET_ISIZE(value, ip)	(ip)->i_size = (value)
543 
544 #else	/* _LP64 */
545 
546 #define	UFS_GET_ISIZE(resultp, ip)				\
547 	{							\
548 		rw_enter(&(ip)->i_contents, RW_READER);		\
549 		*(resultp) = (ip)->i_size;			\
550 		rw_exit(&(ip)->i_contents);			\
551 	}
552 #define	UFS_SET_ISIZE(value, ip)				\
553 	{							\
554 		ASSERT(RW_WRITE_HELD(&(ip)->i_contents));	\
555 		(ip)->i_size = (value);				\
556 	}
557 
558 #endif	/* _LP64 */
559 
560 /*
561  * Allocate the specified block in the inode
562  * and make sure any in-core pages are initialized.
563  */
564 #define	BMAPALLOC(ip, off, size, cr) \
565 	bmap_write((ip), (u_offset_t)(off), (size), BI_NORMAL, NULL, cr)
566 
567 #define	ESAME	(-1)		/* trying to rename linked files (special) */
568 
569 #define	UFS_HOLE	(daddr32_t)-1	/* value used when no block allocated */
570 
571 /*
572  * enums
573  */
574 
575 /* direnter ops */
576 enum de_op { DE_CREATE, DE_MKDIR, DE_LINK, DE_RENAME, DE_SYMLINK, DE_ATTRDIR};
577 
578 /* dirremove ops */
579 enum dr_op { DR_REMOVE, DR_RMDIR, DR_RENAME };
580 
581 /*
582  * block initialization type for bmap_write
583  *
584  * BI_NORMAL - allocate and zero fill pages in memory
585  * BI_ALLOC_ONLY - only allocate the block, do not zero out pages in mem
586  * BI_FALLOCATE - allocate only, do not zero out pages, and store as negative
587  *                block number in inode block list
588  */
589 enum bi_type { BI_NORMAL, BI_ALLOC_ONLY, BI_FALLOCATE };
590 
591 /*
592  * This overlays the fid structure (see vfs.h)
593  *
594  * LP64 note: we use int32_t instead of ino_t since UFS does not use
595  * inode numbers larger than 32-bits and ufid's are passed to NFS
596  * which expects them to not grow in size beyond 10 bytes (12 including
597  * the length).
598  */
599 struct ufid {
600 	ushort_t ufid_len;
601 	ushort_t ufid_flags;
602 	int32_t	ufid_ino;
603 	int32_t	ufid_gen;
604 };
605 
606 /*
607  * each ufs thread (see ufs_thread.c) is managed by this struct
608  */
609 struct ufs_q {
610 	union uq_head {
611 		void		*_uq_generic;	/* first entry on q */
612 		struct inode	*_uq_i;
613 		ufs_failure_t	*_uq_uf;
614 	} _uq_head;
615 	int		uq_ne;		/* # of entries/failures found */
616 	int		uq_lowat;	/* thread runs when ne == lowat */
617 	int		uq_hiwat;	/* synchronous idle if ne >= hiwat */
618 	ushort_t	uq_flags;	/* flags (see below) */
619 	kcondvar_t	uq_cv;		/* for sleep/wakeup */
620 	kthread_id_t	uq_threadp;	/* thread managing this q */
621 	kmutex_t	uq_mutex;	/* protects this struct */
622 };
623 
624 #define	uq_head		_uq_head._uq_generic
625 #define	uq_ihead	_uq_head._uq_i
626 #define	uq_ufhead	_uq_head._uq_uf
627 
628 /*
629  * uq_flags
630  */
631 #define	UQ_EXIT		(0x0001)	/* q server exits at its convenience */
632 #define	UQ_WAIT		(0x0002)	/* thread is waiting on q server */
633 #define	UQ_SUSPEND	(0x0004)	/* request for suspension */
634 #define	UQ_SUSPENDED	(0x0008)	/* thread has suspended itself */
635 #define	UQ_FASTCLIENTS	(0x0010)	/* fast clients in ufs_delq_info */
636 
637 /*
638  * When logging is enabled, statvfs must account for blocks and files that
639  * may be on the delete queue.  Protected by ufsvfsp->vfs_delete.uq_mutex
640  */
641 struct ufs_delq_info {
642 	kcondvar_t	delq_fast_cv;	/* for fast-operating clients */
643 	u_offset_t	delq_unreclaimed_blocks;
644 	ulong_t		delq_unreclaimed_files;
645 };
646 
647 
648 /*
649  * global idle queues
650  * The queues are sized dynamically in proportion to ufs_ninode
651  * which, unless overridden, scales with the amount of memory.
652  * The idle queue is halved whenever it hits the low water mark
653  * (1/4 of ufs_ninode), but can burst to sizes much larger. The number
654  * of hash queues is currently maintained to give on average IQHASHQLEN
655  * entries when the idle queue is at the low water mark.
656  * Note, we do not need to search along the hash queues, but use them
657  * in order to batch together geographically local inodes to allow
658  * their updates (via the log or buffer cache) to require less disk seeks.
659  * This gives an incredible performance boost for logging and a boost for
660  * non logging file systems.
661  */
662 typedef struct {
663 	inode_t *i_chain[2];	/* must match inode_t, but unused */
664 	inode_t *i_freef;	/* must match inode_t, idle list forward */
665 	inode_t *i_freeb;	/* must match inode_t, idle list back  */
666 } iqhead_t;
667 
668 extern struct ufs_q ufs_idle_q;		/* used by global ufs idle thread */
669 extern iqhead_t *ufs_junk_iq;		/* junk idle queues */
670 extern iqhead_t *ufs_useful_iq;		/* useful idle queues */
671 extern int ufs_njunk_iq;		/* number of entries in junk iq */
672 extern int ufs_nuseful_iq;		/* number of entries in useful iq */
673 extern int ufs_niqhash;			/* number of iq hash qs - power of 2 */
674 extern int ufs_iqhashmask;		/* iq hash mask = ufs_niqhash - 1 */
675 
676 #define	IQHASHQLEN 32			/* see comments above */
677 #define	INOCGSHIFT 7			/* 128 inodes per cylinder group */
678 #define	IQHASH(ip) (((ip)->i_number >> INOCGSHIFT) & ufs_iqhashmask)
679 #define	IQNEXT(i) ((i) + 1) & ufs_iqhashmask /* next idle queue */
680 
681 extern struct ufs_q	ufs_hlock;	/* used by global ufs hlock thread */
682 
683 /*
684  * vfs_lfflags flags
685  */
686 #define	UFS_LARGEFILES	((ushort_t)0x1)	/* set if mount allows largefiles */
687 
688 /*
689  * vfs_dfritime flags
690  */
691 #define	UFS_DFRATIME	0x1		/* deferred access time */
692 
693 /*
694  * UFS VFS private data.
695  *
696  * UFS file system instances may be linked on several lists.
697  *
698  * -	The vfs_next field chains together every extant ufs instance; this
699  *	list is rooted at ufs_instances and should be used in preference to
700  *	the overall vfs list (which is properly the province of the generic
701  *	file system code, not of file system implementations).  This same list
702  *	link is used during forcible unmounts to chain together instances that
703  *	can't yet be completely dismantled,
704  *
705  * -	The vfs_wnext field is used within ufs_update to form a work list of
706  *	UFS instances to be synced out.
707  */
708 typedef struct ufsvfs {
709 	struct vfs	*vfs_vfs;	/* back link			*/
710 	struct ufsvfs	*vfs_next;	/* instance list link		*/
711 	struct ufsvfs	*vfs_wnext;	/* work list link		*/
712 	struct vnode	*vfs_root;	/* root vnode			*/
713 	struct buf	*vfs_bufp;	/* buffer containing superblock */
714 	struct vnode	*vfs_devvp;	/* block device vnode		*/
715 	ushort_t	vfs_lfflags;	/* Large files (set by mount)   */
716 	ushort_t	vfs_qflags;	/* QUOTA: filesystem flags	*/
717 	struct inode	*vfs_qinod;	/* QUOTA: pointer to quota file */
718 	uint_t		vfs_btimelimit;	/* QUOTA: block time limit	*/
719 	uint_t		vfs_ftimelimit;	/* QUOTA: file time limit	*/
720 	krwlock_t	vfs_dqrwlock;	/* QUOTA: protects quota fields */
721 	/*
722 	 * some fs local threads
723 	 */
724 	struct ufs_q	vfs_delete;	/* delayed inode delete */
725 	struct ufs_q	vfs_reclaim;	/* reclaim open, deleted files */
726 
727 	/*
728 	 * This is copied from the super block at mount time.
729 	 */
730 	int		vfs_nrpos;	/* # rotational positions */
731 	/*
732 	 * This lock protects cg's and super block pointed at by
733 	 * vfs_bufp->b_fs.  Locks contents of fs and cg's and contents
734 	 * of vfs_dio.
735 	 */
736 	kmutex_t	vfs_lock;
737 	struct ulockfs	vfs_ulockfs;	/* ufs lockfs support */
738 	uint_t		vfs_dio;	/* delayed io (_FIODIO) */
739 	uint_t		vfs_nointr;	/* disallow lockfs interrupts */
740 	uint_t		vfs_nosetsec;	/* disallow ufs_setsecattr */
741 	uint_t		vfs_syncdir;	/* synchronous local directory ops */
742 	uint_t		vfs_dontblock;	/* don't block on forced umount */
743 
744 	/*
745 	 * trans (logging ufs) stuff
746 	 */
747 	uint_t		vfs_domatamap;	/* set if matamap enabled */
748 	ulong_t		vfs_maxacl;	/* transaction stuff - max acl size */
749 	ulong_t		vfs_dirsize;	/* logspace for directory creation */
750 	ulong_t		vfs_avgbfree;	/* average free blks in cg (blkpref) */
751 	/*
752 	 * Some useful constants
753 	 */
754 	int	vfs_nindirshift;	/* calc. from fs_nindir */
755 	int	vfs_nindiroffset;	/* calc. from fs_ninidr */
756 	int	vfs_ioclustsz;		/* bytes in read/write cluster */
757 	int	vfs_iotransz;		/* max device i/o transfer size  */
758 
759 	vfs_ufsfx_t	vfs_fsfx;	/* lock/fix-on-panic support */
760 	/*
761 	 * More useful constants
762 	 */
763 	int	vfs_minfrags;		/* calc. from fs_minfree */
764 	/*
765 	 * Force DirectIO on all files
766 	 */
767 	uint_t	vfs_forcedirectio;
768 	/*
769 	 * Deferred inode time related fields
770 	 */
771 	clock_t		vfs_iotstamp;	/* last I/O timestamp */
772 	uint_t		vfs_dfritime;	/* deferred inode time flags */
773 	/*
774 	 * Some more useful info
775 	 */
776 	dev_t		vfs_dev;	/* device mounted from */
777 	struct ml_unit	*vfs_log;	/* pointer to embedded log struct */
778 	uint_t		vfs_noatime;    /* disable inode atime updates */
779 	/*
780 	 * snapshot stuff
781 	 */
782 	void		*vfs_snapshot;	/* snapshot handle */
783 	/*
784 	 *  Controls logging "file system full" messages to messages file
785 	 */
786 	clock_t		vfs_lastwhinetime;
787 
788 	int 		vfs_nolog_si;	/* not logging summary info */
789 	int		vfs_validfs;	/* indicates mounted fs */
790 
791 	/*
792 	 * Additional information about vfs_delete above
793 	 */
794 	struct ufs_delq_info vfs_delete_info; /* what's on the delete queue */
795 } ufsvfs_t;
796 
797 #define	vfs_fs	vfs_bufp->b_un.b_fs
798 
799 /*
800  * values for vfs_validfs
801  */
802 #define	UT_UNMOUNTED	0
803 #define	UT_MOUNTED	1
804 #define	UT_HLOCKING	2
805 
806 /* inohsz is guaranteed to be a power of 2 */
807 #define	INOHASH(ino)	(((int)ino) & (inohsz - 1))
808 
809 #define	ISFALLOCBLK(ip, bn)	\
810 	(((bn) < 0) && ((bn) % ip->i_fs->fs_frag == 0) && \
811 	((ip)->i_cflags & IFALLOCATE && (bn) != UFS_HOLE))
812 
813 union ihead {
814 	union	ihead	*ih_head[2];
815 	struct	inode	*ih_chain[2];
816 };
817 
818 extern	union	ihead	*ihead;
819 extern  kmutex_t	*ih_lock;
820 extern  int	*ih_ne;
821 extern	int	inohsz;
822 
823 extern	clock_t	ufs_iowait;
824 
825 #endif /* _KERNEL */
826 
827 /*
828  * ufs function prototypes
829  */
830 #if defined(_KERNEL) && !defined(_BOOT)
831 
832 extern	void	ufs_iinit(void);
833 extern	int	ufs_iget(struct vfs *, ino_t, struct inode **, cred_t *);
834 extern	int	ufs_iget_alloced(struct vfs *, ino_t, struct inode **,
835     cred_t *);
836 extern	void	ufs_reset_vnode(vnode_t *);
837 extern	void	ufs_iinactive(struct inode *);
838 extern	void	ufs_iupdat(struct inode *, int);
839 extern	int	ufs_rmidle(struct inode *);
840 extern	int	ufs_itrunc(struct inode *, u_offset_t, int, cred_t *);
841 extern	int	ufs_iaccess(void *, int, cred_t *);
842 extern  int	rdip(struct inode *, struct uio *, int, struct cred *);
843 extern  int	wrip(struct inode *, struct uio *, int, struct cred *);
844 
845 extern void	ufs_imark(struct inode *);
846 extern void	ufs_itimes_nolock(struct inode *);
847 
848 extern	int	ufs_dirlook(struct inode *, char *, struct inode **,
849     cred_t *, int);
850 extern	int	ufs_direnter_cm(struct inode *, char *, enum de_op,
851     struct vattr *, struct inode **, cred_t *, int);
852 extern	int	ufs_direnter_lr(struct inode *, char *, enum de_op,
853     struct inode *, struct inode *, cred_t *, vnode_t **);
854 extern	int	ufs_dircheckpath(ino_t, struct inode *, struct inode *,
855     struct cred *);
856 extern	int	ufs_dirmakeinode(struct inode *, struct inode **,
857     struct vattr *, enum de_op, cred_t *);
858 extern	int	ufs_dirremove(struct inode *, char *, struct inode *,
859     vnode_t *, enum dr_op, cred_t *, vnode_t **);
860 extern  int	ufs_dircheckforname(struct inode *, char *, int,
861     struct ufs_slot *, struct inode **, struct cred *, int);
862 extern	int	ufs_xattrdirempty(struct inode *, ino_t, cred_t *);
863 extern	int	blkatoff(struct inode *, off_t, char **, struct fbuf **);
864 
865 extern	void	sbupdate(struct vfs *);
866 
867 extern	int	ufs_ialloc(struct inode *, ino_t, mode_t, struct inode **,
868     cred_t *);
869 extern	void	ufs_ifree(struct inode *, ino_t, mode_t);
870 extern	void	free(struct inode *, daddr_t, off_t, int);
871 extern	int	alloc(struct inode *, daddr_t, int, daddr_t *, cred_t *);
872 extern	int	realloccg(struct inode *, daddr_t, daddr_t, int, int,
873     daddr_t *, cred_t *);
874 extern	int	ufs_allocsp(struct vnode *, struct flock64 *, cred_t *);
875 extern	int	ufs_freesp(struct vnode *, struct flock64 *, int, cred_t *);
876 extern	ino_t	dirpref(inode_t *);
877 extern	daddr_t	blkpref(struct inode *, daddr_t, int, daddr32_t *);
878 extern	daddr_t	contigpref(ufsvfs_t *, size_t);
879 
880 extern	int	ufs_rdwri(enum uio_rw, int, struct inode *, caddr_t, ssize_t,
881 	offset_t, enum uio_seg, int *, cred_t *);
882 
883 extern	int	bmap_read(struct inode *, u_offset_t, daddr_t *, int *);
884 extern	int	bmap_write(struct inode *, u_offset_t, int, enum bi_type,
885     daddr_t *, struct cred *);
886 extern	int	bmap_has_holes(struct inode *);
887 extern	int	bmap_find(struct inode *, boolean_t, u_offset_t *);
888 extern	int	bmap_set_bn(struct vnode *, u_offset_t, daddr32_t);
889 
890 extern	void	ufs_vfs_add(struct ufsvfs *);
891 extern	void	ufs_vfs_remove(struct ufsvfs *);
892 
893 extern	void	ufs_sbwrite(struct ufsvfs *);
894 extern	void	ufs_update(int);
895 extern	int	ufs_getsummaryinfo(dev_t, struct ufsvfs *, struct fs *);
896 extern	int	ufs_putsummaryinfo(dev_t, struct ufsvfs *, struct fs *);
897 extern	int	ufs_syncip(struct inode *, int, int, top_t);
898 extern	int	ufs_sync_indir(struct inode *);
899 extern	int	ufs_indirblk_sync(struct inode *, offset_t);
900 extern	int	ufs_badblock(struct inode *, daddr_t);
901 extern	int	ufs_indir_badblock(struct inode *, daddr32_t *);
902 extern	void	ufs_notclean(struct ufsvfs *);
903 extern	void	ufs_checkclean(struct vfs *);
904 extern	int	isblock(struct fs *, uchar_t *, daddr_t);
905 extern	void	setblock(struct fs *, uchar_t *, daddr_t);
906 extern	void	clrblock(struct fs *, uchar_t *, daddr_t);
907 extern	int	isclrblock(struct fs *, uchar_t *, daddr_t);
908 extern	void	fragacct(struct fs *, int, int32_t *, int);
909 extern	int	skpc(char, uint_t, char *);
910 extern	int	ufs_fbwrite(struct fbuf *, struct inode *);
911 extern	int	ufs_fbiwrite(struct fbuf *, struct inode *, daddr_t, long);
912 extern	int	ufs_putapage(struct vnode *, struct page *, u_offset_t *,
913 				size_t *, int, struct cred *);
914 extern inode_t	*ufs_alloc_inode(ufsvfs_t *, ino_t);
915 extern void	ufs_free_inode(inode_t *);
916 
917 /*
918  * special stuff
919  */
920 extern	void	ufs_setreclaim(struct inode *);
921 extern	int	ufs_scan_inodes(int, int (*)(struct inode *, void *), void *,
922 				struct ufsvfs *);
923 extern	int	ufs_sync_inode(struct inode *, void *);
924 extern	int	ufs_sticky_remove_access(struct inode *, struct inode *,
925     struct cred *);
926 /*
927  * quota
928  */
929 extern	int	chkiq(struct ufsvfs *, int, struct inode *, uid_t, int,
930 			struct cred *, char **errp, size_t *lenp);
931 
932 /*
933  * ufs thread stuff
934  */
935 extern	void	ufs_thread_delete(struct vfs *);
936 extern	void	ufs_delete_drain(struct vfs *, int, int);
937 extern	void	ufs_delete(struct ufsvfs *, struct inode *, int);
938 extern	void	ufs_inode_cache_reclaim(void *);
939 extern	void	ufs_idle_drain(struct vfs *);
940 extern	void	ufs_idle_some(int);
941 extern	void	ufs_thread_idle(void);
942 extern	void	ufs_thread_reclaim(struct vfs *);
943 extern	void	ufs_thread_init(struct ufs_q *, int);
944 extern	void	ufs_thread_start(struct ufs_q *, void (*)(), struct vfs *);
945 extern	void	ufs_thread_exit(struct ufs_q *);
946 extern	void	ufs_thread_suspend(struct ufs_q *);
947 extern	void	ufs_thread_continue(struct ufs_q *);
948 extern	void	ufs_thread_hlock(void *);
949 extern	void	ufs_delete_init(struct ufsvfs *, int);
950 extern	void	ufs_delete_adjust_stats(struct ufsvfs *, struct statvfs64 *);
951 extern	void	ufs_delete_drain_wait(struct ufsvfs *, int);
952 
953 /*
954  * ufs lockfs stuff
955  */
956 struct seg;
957 extern int ufs_reconcile_fs(struct vfs *, struct ufsvfs *, int);
958 extern int ufs_quiesce(struct ulockfs *);
959 extern int ufs_flush(struct vfs *);
960 extern int ufs_fiolfs(struct vnode *, struct lockfs *, int);
961 extern int ufs__fiolfs(struct vnode *, struct lockfs *, int, int);
962 extern int ufs_fiolfss(struct vnode *, struct lockfs *);
963 extern int ufs_fioffs(struct vnode *, char *, struct cred *);
964 extern int ufs_check_lockfs(struct ufsvfs *, struct ulockfs *, ulong_t);
965 extern int ufs_lockfs_begin(struct ufsvfs *, struct ulockfs **, ulong_t);
966 extern int ufs_lockfs_begin_getpage(struct ufsvfs *, struct ulockfs **,
967 		struct seg *, int, uint_t *);
968 extern void ufs_lockfs_end(struct ulockfs *);
969 /*
970  * ufs acl stuff
971  */
972 extern int ufs_si_inherit(struct inode *, struct inode *, o_mode_t, cred_t *);
973 extern void si_cache_init(void);
974 extern int ufs_si_load(struct inode *, cred_t *);
975 extern void ufs_si_del(struct inode *);
976 extern int ufs_acl_access(struct inode *, int, cred_t *);
977 extern void ufs_si_cache_flush(dev_t);
978 extern int ufs_si_free(si_t *, struct vfs *, cred_t *);
979 extern int ufs_acl_setattr(struct inode *, struct vattr *, cred_t *);
980 extern int ufs_acl_get(struct inode *, vsecattr_t *, int, cred_t *);
981 extern int ufs_acl_set(struct inode *, vsecattr_t *, int, cred_t *);
982 /*
983  * ufs directio stuff
984  */
985 extern void ufs_directio_init();
986 extern int ufs_directio_write(struct inode *, uio_t *, int, int, cred_t *,
987     int *);
988 extern int ufs_directio_read(struct inode *, uio_t *, cred_t *, int *);
989 #define	DIRECTIO_FAILURE	(0)
990 #define	DIRECTIO_SUCCESS	(1)
991 
992 /*
993  * ufs extensions for PXFS
994  */
995 
996 int ufs_rdwr_data(vnode_t *vp, u_offset_t offset, size_t len, fdbuffer_t *fdb,
997     int flags, cred_t *cr);
998 int ufs_alloc_data(vnode_t *vp, u_offset_t offset, size_t *len, fdbuffer_t *fdb,
999     int flags, cred_t *cr);
1000 
1001 /*
1002  * prototypes to support the forced unmount
1003  */
1004 
1005 void ufs_freeze(struct ulockfs *, struct lockfs *);
1006 int ufs_thaw(struct vfs *, struct ufsvfs *, struct ulockfs *);
1007 
1008 /*
1009  * extended attributes
1010  */
1011 
1012 int ufs_xattrmkdir(inode_t *, inode_t **, int, struct cred *);
1013 int ufs_xattr_getattrdir(vnode_t *, inode_t **, int, struct cred *);
1014 void ufs_unhook_shadow(inode_t *, inode_t *);
1015 
1016 #endif	/* defined(_KERNEL) && !defined(_BOOT) */
1017 
1018 #ifdef	__cplusplus
1019 }
1020 #endif
1021 
1022 #endif	/* _SYS_FS_UFS_INODE_H */
1023