xref: /titanic_51/usr/src/uts/common/sys/fs/ufs_inode.h (revision 65908c77dfc02644236ba18bffe67b5ed6f23135)
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 2009 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 #include <sys/isa_defs.h>
43 #include <sys/fbuf.h>
44 #include <sys/fdbuffer.h>
45 #include <sys/fcntl.h>
46 #include <sys/uio.h>
47 #include <sys/t_lock.h>
48 #include <sys/thread.h>
49 #include <sys/cred.h>
50 #include <sys/time.h>
51 #include <sys/types32.h>
52 #include <sys/fs/ufs_fs.h>
53 #include <sys/fs/ufs_lockfs.h>
54 #include <sys/fs/ufs_trans.h>
55 #include <sys/kstat.h>
56 #include <sys/fs/ufs_acl.h>
57 #include <sys/fs/ufs_panic.h>
58 #include <sys/dnlc.h>
59 
60 #ifdef _KERNEL
61 #include <sys/vfs_opreg.h>
62 #endif
63 
64 #ifdef	__cplusplus
65 extern "C" {
66 #endif
67 
68 /*
69  * The I node is the focus of all local file activity in UNIX.
70  * There is a unique inode allocated for each active file,
71  * each current directory, each mounted-on file, each mapping,
72  * and the root.  An inode is `named' by its dev/inumber pair.
73  * Data in icommon is read in from permanent inode on volume.
74  *
75  * Each inode has 5 locks associated with it:
76  *	i_rwlock:	Serializes ufs_write and ufs_setattr request
77  *			and allows ufs_read requests to proceed in parallel.
78  *			Serializes reads/updates to directories.
79  *	vfs_dqrwlock:	Manages quota sub-system quiescence.  See below.
80  *	i_contents:	Protects almost all of the fields in the inode
81  *			except for those listed below. When held
82  *			in writer mode also protects those fields
83  *			listed under i_tlock.
84  *	i_tlock:	When i_tlock is held with the i_contents reader
85  *			lock the i_atime, i_mtime, i_ctime,
86  *			i_delayoff, i_delaylen, i_nextrio, i_writes, i_flag
87  *			i_seq, i_writer & i_mapcnt fields are protected.
88  *			For more i_flag locking info see below.
89  *	ih_lock:	Protects inode hash chain buckets
90  *	ifree_lock:	Protects inode freelist
91  *
92  * Lock ordering:
93  *	i_rwlock > i_contents > i_tlock
94  *	i_rwlock > vfs_dqrwlock > i_contents(writer) > i_tlock
95  *	i_contents > i_tlock
96  *	vfs_dqrwlock > i_contents(writer) > i_tlock
97  *	ih_lock > i_contents > i_tlock
98  *
99  * Making major changes to quota sub-system state, while the file
100  * system is mounted required the addition of another lock.  The
101  * primary lock in the quota sub-system is vfs_dqrwlock in the ufsvfs
102  * structure.  This lock is used to manage quota sub-system quiescence
103  * for a particular file system. Major changes to quota sub-system
104  * state (disabling quotas, enabling quotas, and setting new quota
105  * limits) all require the file system to be quiescent and grabbing
106  * vfs_dqrwlock as writer accomplishes this.  On the other hand,
107  * grabbing vfs_dqrwlock as reader makes the quota sub-system
108  * non-quiescent and lets the quota sub-system know that now is not a
109  * good time to change major quota sub-system state.  Typically
110  * vfs_dqrwlock is grabbed for reading before i_contents is grabbed for
111  * writing.  However, there are cases where vfs_dqrwlock is grabbed for
112  * reading without a corresponding i_contents write grab because there
113  * is no relevant inode.  There are also cases where i_contents is
114  * grabbed for writing when a vfs_dqrwlock read grab is not needed
115  * because the inode changes do not affect quotas.
116  *
117  * Unfortunately, performance considerations have required that we be more
118  * intelligent about using i_tlock when updating i_flag.  Ideally, we would
119  * have simply separated out several of the bits in i_flag into their own
120  * ints to avoid problems.  But, instead, we have implemented the following
121  * rules:
122  *
123  *	o You can update any i_flag field while holding the writer-contents,
124  *	  or by holding the reader-contents AND holding i_tlock.
125  *	  You can only call ITIMES_NOLOCK while holding the writer-contents,
126  *	  or by holding the reader-contents AND holding i_tlock.
127  *
128  *	o For a directory, holding the reader-rw_lock is sufficient for setting
129  *	  IACC.
130  *
131  *	o Races with IREF are avoided by holding the reader contents lock
132  *	  and by holding i_tlock in ufs_rmidle, ufs_putapage, and ufs_getpage.
133  *	  And by holding the writer-contents in ufs_iinactive.
134  *
135  *	o The callers are no longer required to handle the calls to ITIMES
136  *	  and ITIMES_NOLOCK.  The functions that set the i_flag bits are
137  *	  responsible for managing those calls.  The exceptions are the
138  *	  bmap routines.
139  *
140  * SVR4 Extended Fundamental Type (EFT) support:
141  * 	The inode structure has been enhanced to support
142  *	32-bit user-id, 32-bit group-id, and 32-bit device number.
143  *	Standard SVR4 ufs also supports 32-bit mode field.  For the reason
144  *	of backward compatibility with the previous ufs disk format,
145  *	32-bit mode field is not supported.
146  *
147  *	The current inode structure is 100% backward compatible with
148  *	the previous inode structure if no user-id or group-id exceeds
149  *	USHRT_MAX, and no major or minor number of a device number
150  *	stored in an inode exceeds 255.
151  *
152  * Rules for managing i_seq:
153  *	o i_seq is locked under the same rules as i_flag
154  *	o The i_ctime or i_mtime MUST never change without increasing
155  *	  the value of i_seq.
156  *	o You may increase the value of i_seq without the timestamps
157  *	  changing, this may decrease the callers performance but will
158  *	  be functionally correct.
159  *	o The common case is when IUPD or ICHG is set, increase i_seq
160  *	  and immediately call ITIMES* or ufs_iupdat to create a new timestamp.
161  *	o A less common case is the setting of IUPD or ICHG and while still
162  *	  holding the correct lock defer the timestamp and i_seq update
163  *	  until later, but it must still be done before the lock is released.
164  *	  bmap_write is an example of this, where the caller does the update.
165  *	o If multiple changes are being made with the timestamps being
166  *	  updated only at the end, a single increase of i_seq is allowed.
167  *	o If changes are made with IUPD or ICHG being set, but
168  *	  the controlling lock is being dropped before the timestamp is
169  *	  updated, there is a risk that another thread will also change
170  *	  the file, update i_flag, and push just one timestamp update.
171  *	  There is also the risk that another thread calls ITIMES or
172  *	  ufs_iupdat without setting IUPD|ICHG and thus not changing i_seq,
173  *	  this will cause ufs_imark to change the timestamps without changing
174  *	  i_seq. If the controlling lock is dropped, ISEQ must be set to
175  *	  force i_seq to be increased on next ufs_imark, but i_seq MUST still
176  *	  be increased by the original setting thread before its deferred
177  *	  call to ITIMES to insure it is increased the correct number of times.
178  */
179 
180 #define	UID_LONG  (o_uid_t)65535
181 				/* flag value to indicate uid is 32-bit long */
182 #define	GID_LONG  (o_uid_t)65535
183 				/* flag value to indicate gid is 32-bit long */
184 
185 #define	NDADDR	12		/* direct addresses in inode */
186 #define	NIADDR	3		/* indirect addresses in inode */
187 #define	FSL_SIZE (NDADDR + NIADDR - 1) * sizeof (daddr32_t)
188 				/* max fast symbolic name length is 56 */
189 
190 #define	i_fs	i_ufsvfs->vfs_bufp->b_un.b_fs
191 #define	i_vfs	i_vnode->v_vfsp
192 
193 struct 	icommon {
194 	o_mode_t ic_smode;	/*  0: mode and type of file */
195 	short	ic_nlink;	/*  2: number of links to file */
196 	o_uid_t	ic_suid;	/*  4: owner's user id */
197 	o_gid_t	ic_sgid;	/*  6: owner's group id */
198 	u_offset_t ic_lsize;	/*  8: number of bytes in file */
199 #ifdef _KERNEL
200 	struct timeval32 ic_atime;	/* 16: time last accessed */
201 	struct timeval32 ic_mtime;	/* 24: time last modified */
202 	struct timeval32 ic_ctime;	/* 32: last time inode changed */
203 #else
204 	time32_t ic_atime;	/* 16: time last accessed */
205 	int32_t	ic_atspare;
206 	time32_t ic_mtime;	/* 24: time last modified */
207 	int32_t	ic_mtspare;
208 	time32_t ic_ctime;	/* 32: last time inode changed */
209 	int32_t	ic_ctspare;
210 #endif
211 	daddr32_t	ic_db[NDADDR];	/* 40: disk block addresses */
212 	daddr32_t	ic_ib[NIADDR];	/* 88: indirect blocks */
213 	int32_t	ic_flags;	/* 100: cflags */
214 	int32_t	ic_blocks;	/* 104: 512 byte blocks actually held */
215 	int32_t	ic_gen;		/* 108: generation number */
216 	int32_t	ic_shadow;	/* 112: shadow inode */
217 	uid_t	ic_uid;		/* 116: long EFT version of uid */
218 	gid_t	ic_gid;		/* 120: long EFT version of gid */
219 	uint32_t ic_oeftflag;	/* 124: extended attr directory ino, 0 = none */
220 };
221 
222 /*
223  * Large directories can be cached. Directory caching can take the following
224  * states:
225  */
226 typedef enum {
227 	CD_DISABLED_NOMEM = -2,
228 	CD_DISABLED_TOOBIG,
229 	CD_DISABLED,
230 	CD_ENABLED
231 } cachedir_t;
232 
233 /*
234  * Large Files: Note we use the inline functions load_double, store_double
235  * to load and store the long long values of i_size. Therefore the
236  * address of i_size must be eight byte aligned. Kmem_alloc of incore
237  * inode structure makes sure that the structure is 8-byte aligned.
238  * XX64 - reorder this structure?
239  */
240 typedef struct inode {
241 	struct	inode *i_chain[2];	/* must be first */
242 	struct inode *i_freef;	/* free list forward - must be before i_ic */
243 	struct inode *i_freeb;	/* free list back - must be before i_ic */
244 	struct 	icommon	i_ic;	/* Must be here */
245 	struct	vnode *i_vnode;	/* vnode associated with this inode */
246 	struct	vnode *i_devvp;	/* vnode for block I/O */
247 	dev_t	i_dev;		/* device where inode resides */
248 	ino_t	i_number;	/* i number, 1-to-1 with device address */
249 	off_t	i_diroff;	/* offset in dir, where we found last entry */
250 				/* just a hint - no locking needed */
251 	struct ufsvfs *i_ufsvfs; /* incore fs associated with inode */
252 	struct	dquot *i_dquot;	/* quota structure controlling this file */
253 	krwlock_t i_rwlock;	/* serializes write/setattr requests */
254 	krwlock_t i_contents;	/* protects (most of) inode contents */
255 	kmutex_t i_tlock;	/* protects time fields, i_flag */
256 	offset_t i_nextr;	/*					*/
257 				/* next byte read offset (read-ahead)	*/
258 				/*   No lock required			*/
259 				/*					*/
260 	uint_t	i_flag;		/* inode flags */
261 	uint_t	i_seq;		/* modification sequence number */
262 	cachedir_t i_cachedir;	/* Cache this directory on next lookup */
263 				/* - no locking needed  */
264 	long	i_mapcnt;	/* mappings to file pages */
265 	int	*i_map;		/* block list for the corresponding file */
266 	dev_t	i_rdev;		/* INCORE rdev from i_oldrdev by ufs_iget */
267 	size_t	i_delaylen;	/* delayed writes, units=bytes */
268 	offset_t i_delayoff;	/* where we started delaying */
269 	offset_t i_nextrio;	/* where to start the next clust */
270 	long	i_writes;	/* number of outstanding bytes in write q */
271 	kcondvar_t i_wrcv;	/* sleep/wakeup for write throttle */
272 	offset_t i_doff;	/* dinode byte offset in file system */
273 	si_t *i_ufs_acl;	/* pointer to acl entry */
274 	dcanchor_t i_danchor;	/* directory cache anchor */
275 	kthread_t *i_writer;	/* thread which is in window in wrip() */
276 } inode_t;
277 
278 struct dinode {
279 	union {
280 		struct	icommon di_icom;
281 		char	di_size[128];
282 	} di_un;
283 };
284 
285 #define	i_mode		i_ic.ic_smode
286 #define	i_nlink		i_ic.ic_nlink
287 #define	i_uid		i_ic.ic_uid
288 #define	i_gid		i_ic.ic_gid
289 #define	i_smode		i_ic.ic_smode
290 #define	i_suid		i_ic.ic_suid
291 #define	i_sgid		i_ic.ic_sgid
292 
293 #define	i_size		i_ic.ic_lsize
294 #define	i_db		i_ic.ic_db
295 #define	i_ib		i_ic.ic_ib
296 
297 #define	i_atime		i_ic.ic_atime
298 #define	i_mtime		i_ic.ic_mtime
299 #define	i_ctime		i_ic.ic_ctime
300 
301 #define	i_shadow	i_ic.ic_shadow
302 #define	i_oeftflag	i_ic.ic_oeftflag
303 #define	i_blocks	i_ic.ic_blocks
304 #define	i_cflags	i_ic.ic_flags
305 #ifdef _LITTLE_ENDIAN
306 /*
307  * Originally done on x86, but carried on to all other little
308  * architectures, which provides for file system compatibility.
309  */
310 #define	i_ordev		i_ic.ic_db[1]	/* USL SVR4 compatibility */
311 #else
312 #define	i_ordev		i_ic.ic_db[0]	/* was i_oldrdev */
313 #endif
314 #define	i_gen		i_ic.ic_gen
315 #define	i_forw		i_chain[0]
316 #define	i_back		i_chain[1]
317 
318 /* EFT transition aids - obsolete */
319 #define	oEFT_MAGIC	0x90909090
320 #define	di_oeftflag	di_ic.ic_oeftflag
321 
322 #define	di_ic		di_un.di_icom
323 #define	di_mode		di_ic.ic_smode
324 #define	di_nlink	di_ic.ic_nlink
325 #define	di_uid		di_ic.ic_uid
326 #define	di_gid		di_ic.ic_gid
327 #define	di_smode	di_ic.ic_smode
328 #define	di_suid		di_ic.ic_suid
329 #define	di_sgid		di_ic.ic_sgid
330 
331 #define	di_size		di_ic.ic_lsize
332 #define	di_db		di_ic.ic_db
333 #define	di_ib		di_ic.ic_ib
334 
335 #define	di_atime	di_ic.ic_atime
336 #define	di_mtime	di_ic.ic_mtime
337 #define	di_ctime	di_ic.ic_ctime
338 #define	di_cflags	di_ic.ic_flags
339 
340 #ifdef _LITTLE_ENDIAN
341 #define	di_ordev	di_ic.ic_db[1]
342 #else
343 #define	di_ordev	di_ic.ic_db[0]
344 #endif
345 #define	di_shadow	di_ic.ic_shadow
346 #define	di_blocks	di_ic.ic_blocks
347 #define	di_gen		di_ic.ic_gen
348 
349 /* flags */
350 #define	IUPD		0x0001		/* file has been modified */
351 #define	IACC		0x0002		/* inode access time to be updated */
352 #define	IMOD		0x0004		/* inode has been modified */
353 #define	ICHG		0x0008		/* inode has been changed */
354 #define	INOACC		0x0010		/* no access time update in getpage */
355 #define	IMODTIME	0x0020		/* mod time already set */
356 #define	IREF		0x0040		/* inode is being referenced */
357 #define	ISYNC		0x0080		/* do all allocation synchronously */
358 #define	IFASTSYMLNK	0x0100		/* fast symbolic link */
359 #define	IMODACC		0x0200		/* only access time changed; */
360 					/*   filesystem won't become active */
361 #define	IATTCHG		0x0400		/* only size/blocks have changed */
362 #define	IBDWRITE	0x0800		/* the inode has been scheduled for */
363 					/* write operation asynchronously */
364 #define	ISTALE		0x1000		/* inode couldn't be read from disk */
365 #define	IDEL		0x2000		/* inode is being deleted */
366 #define	IDIRECTIO	0x4000		/* attempt directio */
367 #define	ISEQ		0x8000		/* deferred i_seq increase */
368 #define	IJUNKIQ		0x10000		/* on junk idle queue */
369 #define	IQUIET		0x20000		/* No file system full messages */
370 
371 /* cflags */
372 #define	IXATTR		0x0001		/* extended attribute */
373 #define	IFALLOCATE	0x0002		/* fallocate'd file */
374 #define	ICOMPRESS	0x0004		/* compressed for dcfs - see */
375 					/*   `ufs_ioctl()`_FIO_COMPRESSED */
376 
377 /* modes */
378 #define	IFMT		0170000		/* type of file */
379 #define	IFIFO		0010000		/* named pipe (fifo) */
380 #define	IFCHR		0020000		/* character special */
381 #define	IFDIR		0040000		/* directory */
382 #define	IFBLK		0060000		/* block special */
383 #define	IFREG		0100000		/* regular */
384 #define	IFLNK		0120000		/* symbolic link */
385 #define	IFSHAD		0130000		/* shadow indode */
386 #define	IFSOCK		0140000		/* socket */
387 #define	IFATTRDIR	0160000		/* Attribute directory */
388 
389 #define	ISUID		04000		/* set user id on execution */
390 #define	ISGID		02000		/* set group id on execution */
391 #define	ISVTX		01000		/* save swapped text even after use */
392 #define	IREAD		0400		/* read, write, execute permissions */
393 #define	IWRITE		0200
394 #define	IEXEC		0100
395 
396 /* specify how the inode info is written in ufs_syncip() */
397 #define	I_SYNC		1		/* wait for the inode written to disk */
398 #define	I_DSYNC		2		/* wait for the inode written to disk */
399 					/* only if IATTCHG is set */
400 #define	I_ASYNC		0		/* don't wait for the inode written */
401 
402 /* flags passed to ufs_itrunc(), indirtrunc(), and free() */
403 #define	I_FREE	0x00000001		/* inode is being freed */
404 #define	I_DIR	0x00000002		/* inode is a directory */
405 #define	I_IBLK	0x00000004		/* indirect block */
406 #define	I_CHEAP	0x00000008		/* cheap free */
407 #define	I_SHAD	0x00000010		/* inode is a shadow inode */
408 #define	I_QUOTA	0x00000020		/* quota file */
409 #define	I_NOCANCEL	0x40		/* Don't cancel these fragments */
410 #define	I_ACCT	0x00000080		/* Update ufsvfs' unreclaimed_blocks */
411 
412 /*
413  * If ufs_dircheckforname() fails to find an entry with the given name,
414  * this "slot" structure holds state for ufs_direnter_*() as to where
415  * there is space to put an entry with that name.
416  * If ufs_dircheckforname() finds an entry with the given name, this structure
417  * holds state for ufs_dirrename() and ufs_dirremove() as to where the
418  * entry is. "status" indicates what ufs_dircheckforname() found:
419  *      NONE            name not found, large enough free slot not found,
420  *      FOUND           name not found, large enough free slot found
421  *      EXIST           name found
422  * If ufs_dircheckforname() fails due to an error, this structure is not
423  * filled in.
424  *
425  * After ufs_dircheckforname() succeeds the values are:
426  *      status  offset          size            fbp, ep
427  *      ------  ------          ----            -------
428  *      NONE    end of dir      needed          not valid
429  *      FOUND   start of entry  of ent          both valid if fbp != NULL
430  *      EXIST   start of entry  of prev ent     valid
431  *
432  * "endoff" is set to 0 if the an entry with the given name is found, or if no
433  * free slot could be found or made; this means that the directory should not
434  * be truncated.  If the entry was found, the search terminates so
435  * ufs_dircheckforname() didn't find out where the last valid entry in the
436  * directory was, so it doesn't know where to cut the directory off; if no free
437  * slot could be found or made, the directory has to be extended to make room
438  * for the new entry, so there's nothing to cut off.
439  * Otherwise, "endoff" is set to the larger of the offset of the last
440  * non-empty entry in the directory, or the offset at which the new entry will
441  * be placed, whichever is larger.  This is used by ufs_diraddentry(); if a new
442  * entry is to be added to the directory, any complete directory blocks at the
443  * end of the directory that contain no non-empty entries are lopped off the
444  * end, thus shrinking the directory dynamically.
445  */
446 typedef enum {NONE, FOUND, EXIST} slotstat_t;
447 struct ufs_slot {
448 	struct	direct *ep;	/* pointer to slot */
449 	struct	fbuf *fbp;	/* dir buf where slot is */
450 	off_t	offset;		/* offset of area with free space */
451 	off_t	endoff;		/* last useful location found in search */
452 	slotstat_t status;	/* status of slot */
453 	int	size;		/* size of area at slotoffset */
454 	int	cached;		/* cached directory */
455 };
456 
457 /*
458  * Statistics on inodes
459  * Not protected by locks
460  */
461 struct instats {
462 	kstat_named_t in_size;		/* current cache size */
463 	kstat_named_t in_maxsize;	/* maximum cache size */
464 	kstat_named_t in_hits;		/* cache hits */
465 	kstat_named_t in_misses;	/* cache misses */
466 	kstat_named_t in_malloc;	/* kmem_alloce'd */
467 	kstat_named_t in_mfree;		/* kmem_free'd */
468 	kstat_named_t in_maxreached;	/* Largest size reached by cache */
469 	kstat_named_t in_frfront;	/* # put at front of freelist */
470 	kstat_named_t in_frback;	/* # put at back of freelist */
471 	kstat_named_t in_qfree;		/* q's to delete thread */
472 	kstat_named_t in_scan;		/* # inodes scanned */
473 	kstat_named_t in_tidles;	/* # inodes idled by idle thread */
474 	kstat_named_t in_lidles;	/* # inodes idled by ufs_lookup */
475 	kstat_named_t in_vidles;	/* # inodes idled by ufs_vget */
476 	kstat_named_t in_kcalloc;	/* # inodes kmem_cache_alloced */
477 	kstat_named_t in_kcfree;	/* # inodes kmem_cache_freed */
478 	kstat_named_t in_poc;		/* # push-on-close's */
479 };
480 
481 #ifdef _KERNEL
482 
483 /*
484  * Extended attributes
485  */
486 
487 #define	XATTR_DIR_NAME	"/@/"
488 extern int	ufs_ninode;		/* high-water mark for inode cache */
489 
490 extern struct vnodeops *ufs_vnodeops;	/* vnode operations for ufs */
491 extern const struct fs_operation_def ufs_vnodeops_template[];
492 
493 /*
494  * Convert between inode pointers and vnode pointers
495  */
496 #define	VTOI(VP)	((struct inode *)(VP)->v_data)
497 #define	ITOV(IP)	((struct vnode *)(IP)->i_vnode)
498 
499 /*
500  * convert to fs
501  */
502 #define	ITOF(IP)	((struct fs *)(IP)->i_fs)
503 
504 /*
505  * Convert between vnode types and inode formats
506  */
507 extern enum vtype	iftovt_tab[];
508 
509 #ifdef notneeded
510 
511 /* Look at sys/mode.h and os/vnode.c */
512 
513 extern int		vttoif_tab[];
514 
515 #endif
516 
517 /*
518  * Mark an inode with the current (unique) timestamp.
519  * (Note that UFS's concept of time only keeps 32 bits of seconds
520  * in the on-disk format).
521  */
522 struct timeval32 iuniqtime;
523 extern kmutex_t ufs_iuniqtime_lock;
524 
525 #define	ITIMES_NOLOCK(ip) ufs_itimes_nolock(ip)
526 
527 #define	ITIMES(ip) { \
528 	mutex_enter(&(ip)->i_tlock); \
529 	ITIMES_NOLOCK(ip); \
530 	mutex_exit(&(ip)->i_tlock); \
531 }
532 
533 /*
534  * The following interfaces are used to do atomic loads and stores
535  * of an inode's i_size, which is a long long data type.
536  *
537  * For LP64, we just to a load or a store - atomicity and alignment
538  * are 8-byte guaranteed.  For x86 there are no such instructions,
539  * so we grab i_contents as reader to get the size; we already hold
540  * it as writer when we're setting the size.
541  */
542 
543 #ifdef _LP64
544 
545 #define	UFS_GET_ISIZE(resultp, ip)	*(resultp) = (ip)->i_size
546 #define	UFS_SET_ISIZE(value, ip)	(ip)->i_size = (value)
547 
548 #else	/* _LP64 */
549 
550 #define	UFS_GET_ISIZE(resultp, ip)				\
551 	{							\
552 		rw_enter(&(ip)->i_contents, RW_READER);		\
553 		*(resultp) = (ip)->i_size;			\
554 		rw_exit(&(ip)->i_contents);			\
555 	}
556 #define	UFS_SET_ISIZE(value, ip)				\
557 	{							\
558 		ASSERT(RW_WRITE_HELD(&(ip)->i_contents));	\
559 		(ip)->i_size = (value);				\
560 	}
561 
562 #endif	/* _LP64 */
563 
564 /*
565  * Allocate the specified block in the inode
566  * and make sure any in-core pages are initialized.
567  */
568 #define	BMAPALLOC(ip, off, size, cr) \
569 	bmap_write((ip), (u_offset_t)(off), (size), BI_NORMAL, NULL, cr)
570 
571 #define	ESAME	(-1)		/* trying to rename linked files (special) */
572 
573 #define	UFS_HOLE	(daddr32_t)-1	/* value used when no block allocated */
574 
575 /*
576  * enums
577  */
578 
579 /* direnter ops */
580 enum de_op { DE_CREATE, DE_MKDIR, DE_LINK, DE_RENAME, DE_SYMLINK, DE_ATTRDIR};
581 
582 /* dirremove ops */
583 enum dr_op { DR_REMOVE, DR_RMDIR, DR_RENAME };
584 
585 /*
586  * block initialization type for bmap_write
587  *
588  * BI_NORMAL - allocate and zero fill pages in memory
589  * BI_ALLOC_ONLY - only allocate the block, do not zero out pages in mem
590  * BI_FALLOCATE - allocate only, do not zero out pages, and store as negative
591  *                block number in inode block list
592  */
593 enum bi_type { BI_NORMAL, BI_ALLOC_ONLY, BI_FALLOCATE };
594 
595 /*
596  * This overlays the fid structure (see vfs.h)
597  *
598  * LP64 note: we use int32_t instead of ino_t since UFS does not use
599  * inode numbers larger than 32-bits and ufid's are passed to NFS
600  * which expects them to not grow in size beyond 10 bytes (12 including
601  * the length).
602  */
603 struct ufid {
604 	ushort_t ufid_len;
605 	ushort_t ufid_flags;
606 	int32_t	ufid_ino;
607 	int32_t	ufid_gen;
608 };
609 
610 /*
611  * each ufs thread (see ufs_thread.c) is managed by this struct
612  */
613 struct ufs_q {
614 	union uq_head {
615 		void		*_uq_generic;	/* first entry on q */
616 		struct inode	*_uq_i;
617 		ufs_failure_t	*_uq_uf;
618 	} _uq_head;
619 	int		uq_ne;		/* # of entries/failures found */
620 	int		uq_lowat;	/* thread runs when ne == lowat */
621 	int		uq_hiwat;	/* synchronous idle if ne >= hiwat */
622 	ushort_t	uq_flags;	/* flags (see below) */
623 	kcondvar_t	uq_cv;		/* for sleep/wakeup */
624 	kthread_id_t	uq_threadp;	/* thread managing this q */
625 	kmutex_t	uq_mutex;	/* protects this struct */
626 };
627 
628 #define	uq_head		_uq_head._uq_generic
629 #define	uq_ihead	_uq_head._uq_i
630 #define	uq_ufhead	_uq_head._uq_uf
631 
632 /*
633  * uq_flags
634  */
635 #define	UQ_EXIT		(0x0001)	/* q server exits at its convenience */
636 #define	UQ_WAIT		(0x0002)	/* thread is waiting on q server */
637 #define	UQ_SUSPEND	(0x0004)	/* request for suspension */
638 #define	UQ_SUSPENDED	(0x0008)	/* thread has suspended itself */
639 
640 /*
641  * When logging is enabled, statvfs must account for blocks and files that
642  * may be on the delete queue.  Protected by ufsvfsp->vfs_delete.uq_mutex
643  */
644 struct ufs_delq_info {
645 	u_offset_t	delq_unreclaimed_blocks;
646 	ulong_t		delq_unreclaimed_files;
647 };
648 
649 
650 /*
651  * global idle queues
652  * The queues are sized dynamically in proportion to ufs_ninode
653  * which, unless overridden, scales with the amount of memory.
654  * The idle queue is halved whenever it hits the low water mark
655  * (1/4 of ufs_ninode), but can burst to sizes much larger. The number
656  * of hash queues is currently maintained to give on average IQHASHQLEN
657  * entries when the idle queue is at the low water mark.
658  * Note, we do not need to search along the hash queues, but use them
659  * in order to batch together geographically local inodes to allow
660  * their updates (via the log or buffer cache) to require less disk seeks.
661  * This gives an incredible performance boost for logging and a boost for
662  * non logging file systems.
663  */
664 typedef struct {
665 	inode_t *i_chain[2];	/* must match inode_t, but unused */
666 	inode_t *i_freef;	/* must match inode_t, idle list forward */
667 	inode_t *i_freeb;	/* must match inode_t, idle list back  */
668 } iqhead_t;
669 
670 extern struct ufs_q ufs_idle_q;		/* used by global ufs idle thread */
671 extern iqhead_t *ufs_junk_iq;		/* junk idle queues */
672 extern iqhead_t *ufs_useful_iq;		/* useful idle queues */
673 extern int ufs_njunk_iq;		/* number of entries in junk iq */
674 extern int ufs_nuseful_iq;		/* number of entries in useful iq */
675 extern int ufs_niqhash;			/* number of iq hash qs - power of 2 */
676 extern int ufs_iqhashmask;		/* iq hash mask = ufs_niqhash - 1 */
677 
678 #define	IQHASHQLEN 32			/* see comments above */
679 #define	INOCGSHIFT 7			/* 128 inodes per cylinder group */
680 #define	IQHASH(ip) (((ip)->i_number >> INOCGSHIFT) & ufs_iqhashmask)
681 #define	IQNEXT(i) ((i) + 1) & ufs_iqhashmask /* next idle queue */
682 
683 extern struct ufs_q	ufs_hlock;	/* used by global ufs hlock thread */
684 
685 /*
686  * vfs_lfflags flags
687  */
688 #define	UFS_LARGEFILES	((ushort_t)0x1)	/* set if mount allows largefiles */
689 
690 /*
691  * vfs_dfritime flags
692  */
693 #define	UFS_DFRATIME	0x1		/* deferred access time */
694 
695 /*
696  * UFS VFS private data.
697  *
698  * UFS file system instances may be linked on several lists.
699  *
700  * -	The vfs_next field chains together every extant ufs instance; this
701  *	list is rooted at ufs_instances and should be used in preference to
702  *	the overall vfs list (which is properly the province of the generic
703  *	file system code, not of file system implementations).  This same list
704  *	link is used during forcible unmounts to chain together instances that
705  *	can't yet be completely dismantled,
706  *
707  * -	The vfs_wnext field is used within ufs_update to form a work list of
708  *	UFS instances to be synced out.
709  */
710 typedef struct ufsvfs {
711 	struct vfs	*vfs_vfs;	/* back link			*/
712 	struct ufsvfs	*vfs_next;	/* instance list link		*/
713 	struct ufsvfs	*vfs_wnext;	/* work list link		*/
714 	struct vnode	*vfs_root;	/* root vnode			*/
715 	struct buf	*vfs_bufp;	/* buffer containing superblock */
716 	struct vnode	*vfs_devvp;	/* block device vnode		*/
717 	ushort_t	vfs_lfflags;	/* Large files (set by mount)   */
718 	ushort_t	vfs_qflags;	/* QUOTA: filesystem flags	*/
719 	struct inode	*vfs_qinod;	/* QUOTA: pointer to quota file */
720 	uint_t		vfs_btimelimit;	/* QUOTA: block time limit	*/
721 	uint_t		vfs_ftimelimit;	/* QUOTA: file time limit	*/
722 	krwlock_t	vfs_dqrwlock;	/* QUOTA: protects quota fields */
723 	/*
724 	 * some fs local threads
725 	 */
726 	struct ufs_q	vfs_delete;	/* delayed inode delete */
727 	struct ufs_q	vfs_reclaim;	/* reclaim open, deleted files */
728 
729 	/*
730 	 * This is copied from the super block at mount time.
731 	 */
732 	int		vfs_nrpos;	/* # rotational positions */
733 	/*
734 	 * This lock protects cg's and super block pointed at by
735 	 * vfs_bufp->b_fs.  Locks contents of fs and cg's and contents
736 	 * of vfs_dio.
737 	 */
738 	kmutex_t	vfs_lock;
739 	struct ulockfs	vfs_ulockfs;	/* ufs lockfs support */
740 	uint_t		vfs_dio;	/* delayed io (_FIODIO) */
741 	uint_t		vfs_nointr;	/* disallow lockfs interrupts */
742 	uint_t		vfs_nosetsec;	/* disallow ufs_setsecattr */
743 	uint_t		vfs_syncdir;	/* synchronous local directory ops */
744 	uint_t		vfs_dontblock;	/* don't block on forced umount */
745 
746 	/*
747 	 * trans (logging ufs) stuff
748 	 */
749 	uint_t		vfs_domatamap;	/* set if matamap enabled */
750 	ulong_t		vfs_maxacl;	/* transaction stuff - max acl size */
751 	ulong_t		vfs_dirsize;	/* logspace for directory creation */
752 	ulong_t		vfs_avgbfree;	/* average free blks in cg (blkpref) */
753 	/*
754 	 * Some useful constants
755 	 */
756 	int	vfs_nindirshift;	/* calc. from fs_nindir */
757 	int	vfs_nindiroffset;	/* calc. from fs_ninidr */
758 	int	vfs_ioclustsz;		/* bytes in read/write cluster */
759 	int	vfs_iotransz;		/* max device i/o transfer size  */
760 
761 	vfs_ufsfx_t	vfs_fsfx;	/* lock/fix-on-panic support */
762 	/*
763 	 * More useful constants
764 	 */
765 	int	vfs_minfrags;		/* calc. from fs_minfree */
766 	/*
767 	 * Force DirectIO on all files
768 	 */
769 	uint_t	vfs_forcedirectio;
770 	/*
771 	 * Deferred inode time related fields
772 	 */
773 	clock_t		vfs_iotstamp;	/* last I/O timestamp */
774 	uint_t		vfs_dfritime;	/* deferred inode time flags */
775 	/*
776 	 * Some more useful info
777 	 */
778 	dev_t		vfs_dev;	/* device mounted from */
779 	struct ml_unit	*vfs_log;	/* pointer to embedded log struct */
780 	uint_t		vfs_noatime;    /* disable inode atime updates */
781 	/*
782 	 * snapshot stuff
783 	 */
784 	void		*vfs_snapshot;	/* snapshot handle */
785 	/*
786 	 *  Controls logging "file system full" messages to messages file
787 	 */
788 	clock_t		vfs_lastwhinetime;
789 
790 	int 		vfs_nolog_si;	/* not logging summary info */
791 	int		vfs_validfs;	/* indicates mounted fs */
792 
793 	/*
794 	 * Additional information about vfs_delete above
795 	 */
796 	struct ufs_delq_info vfs_delete_info; /* what's on the delete queue */
797 } ufsvfs_t;
798 
799 #define	vfs_fs	vfs_bufp->b_un.b_fs
800 
801 /*
802  * values for vfs_validfs
803  */
804 #define	UT_UNMOUNTED	0
805 #define	UT_MOUNTED	1
806 #define	UT_HLOCKING	2
807 
808 /* inohsz is guaranteed to be a power of 2 */
809 #define	INOHASH(ino)	(((int)ino) & (inohsz - 1))
810 
811 #define	ISFALLOCBLK(ip, bn)	\
812 	(((bn) < 0) && ((bn) % ip->i_fs->fs_frag == 0) && \
813 	((ip)->i_cflags & IFALLOCATE && (bn) != UFS_HOLE))
814 
815 union ihead {
816 	union	ihead	*ih_head[2];
817 	struct	inode	*ih_chain[2];
818 };
819 
820 extern	union	ihead	*ihead;
821 extern  kmutex_t	*ih_lock;
822 extern  int	*ih_ne;
823 extern	int	inohsz;
824 
825 extern	clock_t	ufs_iowait;
826 
827 #endif /* _KERNEL */
828 
829 /*
830  * ufs function prototypes
831  */
832 #if defined(_KERNEL) && !defined(_BOOT)
833 
834 extern	void	ufs_iinit(void);
835 extern	int	ufs_iget(struct vfs *, ino_t, struct inode **, cred_t *);
836 extern	int	ufs_iget_alloced(struct vfs *, ino_t, struct inode **,
837     cred_t *);
838 extern	void	ufs_reset_vnode(vnode_t *);
839 extern	void	ufs_iinactive(struct inode *);
840 extern	void	ufs_iupdat(struct inode *, int);
841 extern	int	ufs_rmidle(struct inode *);
842 extern	int	ufs_itrunc(struct inode *, u_offset_t, int, cred_t *);
843 extern	int	ufs_iaccess(struct inode *, int, cred_t *, int);
844 extern  int	rdip(struct inode *, struct uio *, int, struct cred *);
845 extern  int	wrip(struct inode *, struct uio *, int, struct cred *);
846 
847 extern void	ufs_imark(struct inode *);
848 extern void	ufs_itimes_nolock(struct inode *);
849 
850 extern	int	ufs_diraccess(struct inode *, int, struct cred *);
851 extern	int	ufs_dirlook(struct inode *, char *, struct inode **,
852     cred_t *, int);
853 extern	int	ufs_direnter_cm(struct inode *, char *, enum de_op,
854     struct vattr *, struct inode **, cred_t *, int);
855 extern	int	ufs_direnter_lr(struct inode *, char *, enum de_op,
856     struct inode *, struct inode *, cred_t *, vnode_t **);
857 extern	int	ufs_dircheckpath(ino_t, struct inode *, struct inode *,
858     struct cred *);
859 extern	int	ufs_dirmakeinode(struct inode *, struct inode **,
860     struct vattr *, enum de_op, cred_t *);
861 extern	int	ufs_dirremove(struct inode *, char *, struct inode *,
862     vnode_t *, enum dr_op, cred_t *, vnode_t **);
863 extern  int	ufs_dircheckforname(struct inode *, char *, int,
864     struct ufs_slot *, struct inode **, struct cred *, int);
865 extern	int	ufs_xattrdirempty(struct inode *, ino_t, cred_t *);
866 extern	int	blkatoff(struct inode *, off_t, char **, struct fbuf **);
867 
868 extern	void	sbupdate(struct vfs *);
869 
870 extern	int	ufs_ialloc(struct inode *, ino_t, mode_t, struct inode **,
871     cred_t *);
872 extern	void	ufs_ifree(struct inode *, ino_t, mode_t);
873 extern	void	free(struct inode *, daddr_t, off_t, int);
874 extern	int	alloc(struct inode *, daddr_t, int, daddr_t *, cred_t *);
875 extern	int	realloccg(struct inode *, daddr_t, daddr_t, int, int,
876     daddr_t *, cred_t *);
877 extern	int	ufs_allocsp(struct vnode *, struct flock64 *, cred_t *);
878 extern	int	ufs_freesp(struct vnode *, struct flock64 *, int, cred_t *);
879 extern	ino_t	dirpref(inode_t *);
880 extern	daddr_t	blkpref(struct inode *, daddr_t, int, daddr32_t *);
881 extern	daddr_t	contigpref(ufsvfs_t *, size_t);
882 
883 extern	int	ufs_rdwri(enum uio_rw, int, struct inode *, caddr_t, ssize_t,
884 	offset_t, enum uio_seg, int *, cred_t *);
885 
886 extern	int	bmap_read(struct inode *, u_offset_t, daddr_t *, int *);
887 extern	int	bmap_write(struct inode *, u_offset_t, int, enum bi_type,
888     daddr_t *, struct cred *);
889 extern	int	bmap_has_holes(struct inode *);
890 extern	int	bmap_find(struct inode *, boolean_t, u_offset_t *);
891 extern	int	bmap_set_bn(struct vnode *, u_offset_t, daddr32_t);
892 
893 extern	void	ufs_vfs_add(struct ufsvfs *);
894 extern	void	ufs_vfs_remove(struct ufsvfs *);
895 
896 extern	void	ufs_sbwrite(struct ufsvfs *);
897 extern	void	ufs_update(int);
898 extern	int	ufs_getsummaryinfo(dev_t, struct ufsvfs *, struct fs *);
899 extern	int	ufs_putsummaryinfo(dev_t, struct ufsvfs *, struct fs *);
900 extern	int	ufs_syncip(struct inode *, int, int, top_t);
901 extern	int	ufs_sync_indir(struct inode *);
902 extern	int	ufs_indirblk_sync(struct inode *, offset_t);
903 extern	int	ufs_badblock(struct inode *, daddr_t);
904 extern	int	ufs_indir_badblock(struct inode *, daddr32_t *);
905 extern	void	ufs_notclean(struct ufsvfs *);
906 extern	void	ufs_checkclean(struct vfs *);
907 extern	int	isblock(struct fs *, uchar_t *, daddr_t);
908 extern	void	setblock(struct fs *, uchar_t *, daddr_t);
909 extern	void	clrblock(struct fs *, uchar_t *, daddr_t);
910 extern	int	isclrblock(struct fs *, uchar_t *, daddr_t);
911 extern	void	fragacct(struct fs *, int, int32_t *, int);
912 extern	int	skpc(char, uint_t, char *);
913 extern	int	ufs_fbwrite(struct fbuf *, struct inode *);
914 extern	int	ufs_fbiwrite(struct fbuf *, struct inode *, daddr_t, long);
915 extern	int	ufs_putapage(struct vnode *, struct page *, u_offset_t *,
916 				size_t *, int, struct cred *);
917 extern inode_t	*ufs_alloc_inode(ufsvfs_t *, ino_t);
918 extern void	ufs_free_inode(inode_t *);
919 
920 /*
921  * special stuff
922  */
923 extern	void	ufs_setreclaim(struct inode *);
924 extern	int	ufs_scan_inodes(int, int (*)(struct inode *, void *), void *,
925 				struct ufsvfs *);
926 extern	int	ufs_sync_inode(struct inode *, void *);
927 extern	int	ufs_sticky_remove_access(struct inode *, struct inode *,
928     struct cred *);
929 /*
930  * quota
931  */
932 extern	int	chkiq(struct ufsvfs *, int, struct inode *, uid_t, int,
933 			struct cred *, char **errp, size_t *lenp);
934 
935 /*
936  * ufs thread stuff
937  */
938 extern	void	ufs_thread_delete(struct vfs *);
939 extern	void	ufs_delete_drain(struct vfs *, int, int);
940 extern	void	ufs_delete(struct ufsvfs *, struct inode *, int);
941 extern	void	ufs_inode_cache_reclaim(void *);
942 extern	void	ufs_idle_drain(struct vfs *);
943 extern	void	ufs_idle_some(int);
944 extern	void	ufs_thread_idle(void);
945 extern	void	ufs_thread_reclaim(struct vfs *);
946 extern	void	ufs_thread_init(struct ufs_q *, int);
947 extern	void	ufs_thread_start(struct ufs_q *, void (*)(), struct vfs *);
948 extern	void	ufs_thread_exit(struct ufs_q *);
949 extern	void	ufs_thread_suspend(struct ufs_q *);
950 extern	void	ufs_thread_continue(struct ufs_q *);
951 extern	void	ufs_thread_hlock(void *);
952 extern	void	ufs_delete_init(struct ufsvfs *, int);
953 extern	void	ufs_delete_adjust_stats(struct ufsvfs *, struct statvfs64 *);
954 extern	void	ufs_delete_drain_wait(struct ufsvfs *, int);
955 
956 /*
957  * ufs lockfs stuff
958  */
959 struct seg;
960 extern int ufs_reconcile_fs(struct vfs *, struct ufsvfs *, int);
961 extern int ufs_quiesce(struct ulockfs *);
962 extern int ufs_flush(struct vfs *);
963 extern int ufs_fiolfs(struct vnode *, struct lockfs *, int);
964 extern int ufs__fiolfs(struct vnode *, struct lockfs *, int, int);
965 extern int ufs_fiolfss(struct vnode *, struct lockfs *);
966 extern int ufs_fioffs(struct vnode *, char *, struct cred *);
967 extern int ufs_check_lockfs(struct ufsvfs *, struct ulockfs *, ulong_t);
968 extern int ufs_lockfs_begin(struct ufsvfs *, struct ulockfs **, ulong_t);
969 extern int ufs_lockfs_trybegin(struct ufsvfs *, struct ulockfs **, ulong_t);
970 extern int ufs_lockfs_begin_getpage(struct ufsvfs *, struct ulockfs **,
971 		struct seg *, int, uint_t *);
972 extern void ufs_lockfs_end(struct ulockfs *);
973 /*
974  * ufs acl stuff
975  */
976 extern int ufs_si_inherit(struct inode *, struct inode *, o_mode_t, cred_t *);
977 extern void si_cache_init(void);
978 extern int ufs_si_load(struct inode *, cred_t *);
979 extern void ufs_si_del(struct inode *);
980 extern int ufs_acl_access(struct inode *, int, cred_t *);
981 extern void ufs_si_cache_flush(dev_t);
982 extern int ufs_si_free(si_t *, struct vfs *, cred_t *);
983 extern int ufs_acl_setattr(struct inode *, struct vattr *, cred_t *);
984 extern int ufs_acl_get(struct inode *, vsecattr_t *, int, cred_t *);
985 extern int ufs_acl_set(struct inode *, vsecattr_t *, int, cred_t *);
986 /*
987  * ufs directio stuff
988  */
989 extern void ufs_directio_init();
990 extern int ufs_directio_write(struct inode *, uio_t *, int, int, cred_t *,
991     int *);
992 extern int ufs_directio_read(struct inode *, uio_t *, cred_t *, int *);
993 #define	DIRECTIO_FAILURE	(0)
994 #define	DIRECTIO_SUCCESS	(1)
995 
996 /*
997  * ufs extensions for PXFS
998  */
999 
1000 int ufs_rdwr_data(vnode_t *vp, u_offset_t offset, size_t len, fdbuffer_t *fdb,
1001     int flags, cred_t *cr);
1002 int ufs_alloc_data(vnode_t *vp, u_offset_t offset, size_t *len, fdbuffer_t *fdb,
1003     int flags, cred_t *cr);
1004 
1005 /*
1006  * prototypes to support the forced unmount
1007  */
1008 
1009 void ufs_freeze(struct ulockfs *, struct lockfs *);
1010 int ufs_thaw(struct vfs *, struct ufsvfs *, struct ulockfs *);
1011 
1012 /*
1013  * extended attributes
1014  */
1015 
1016 int ufs_xattrmkdir(inode_t *, inode_t **, int, struct cred *);
1017 int ufs_xattr_getattrdir(vnode_t *, inode_t **, int, struct cred *);
1018 void ufs_unhook_shadow(inode_t *, inode_t *);
1019 
1020 #endif	/* defined(_KERNEL) && !defined(_BOOT) */
1021 
1022 #ifdef	__cplusplus
1023 }
1024 #endif
1025 
1026 #endif	/* _SYS_FS_UFS_INODE_H */
1027