master
   1/*-
   2 * SPDX-License-Identifier: BSD-2-Clause
   3 *
   4 * Copyright 1998, 2000 Marshall Kirk McKusick. All Rights Reserved.
   5 *
   6 * The soft updates code is derived from the appendix of a University
   7 * of Michigan technical report (Gregory R. Ganger and Yale N. Patt,
   8 * "Soft Updates: A Solution to the Metadata Update Problem in File
   9 * Systems", CSE-TR-254-95, August 1995).
  10 *
  11 * Further information about soft updates can be obtained from:
  12 *
  13 *	Marshall Kirk McKusick		http://www.mckusick.com/softdep/
  14 *	1614 Oxford Street		mckusick@mckusick.com
  15 *	Berkeley, CA 94709-1608		+1-510-843-9542
  16 *	USA
  17 *
  18 * Redistribution and use in source and binary forms, with or without
  19 * modification, are permitted provided that the following conditions
  20 * are met:
  21 *
  22 * 1. Redistributions of source code must retain the above copyright
  23 *    notice, this list of conditions and the following disclaimer.
  24 * 2. Redistributions in binary form must reproduce the above copyright
  25 *    notice, this list of conditions and the following disclaimer in the
  26 *    documentation and/or other materials provided with the distribution.
  27 *
  28 * THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``AS IS'' AND ANY
  29 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
  30 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  31 * DISCLAIMED.  IN NO EVENT SHALL MARSHALL KIRK MCKUSICK BE LIABLE FOR
  32 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  38 * SUCH DAMAGE.
  39 *
  40 *	@(#)softdep.h	9.7 (McKusick) 6/21/00
  41 */
  42
  43#include <sys/queue.h>
  44
  45/*
  46 * Allocation dependencies are handled with undo/redo on the in-memory
  47 * copy of the data. A particular data dependency is eliminated when
  48 * it is ALLCOMPLETE: that is ATTACHED, DEPCOMPLETE, and COMPLETE.
  49 * 
  50 * The ATTACHED flag means that the data is not currently being written
  51 * to disk.
  52 * 
  53 * The UNDONE flag means that the data has been rolled back to a safe
  54 * state for writing to the disk. When the I/O completes, the data is
  55 * restored to its current form and the state reverts to ATTACHED.
  56 * The data must be locked throughout the rollback, I/O, and roll
  57 * forward so that the rolled back information is never visible to
  58 * user processes.
  59 *
  60 * The COMPLETE flag indicates that the item has been written. For example,
  61 * a dependency that requires that an inode be written will be marked
  62 * COMPLETE after the inode has been written to disk.
  63 * 
  64 * The DEPCOMPLETE flag indicates the completion of any other
  65 * dependencies such as the writing of a cylinder group map has been
  66 * completed. A dependency structure may be freed only when both it
  67 * and its dependencies have completed and any rollbacks that are in
  68 * progress have finished as indicated by the set of ALLCOMPLETE flags
  69 * all being set.
  70 * 
  71 * The two MKDIR flags indicate additional dependencies that must be done
  72 * when creating a new directory. MKDIR_BODY is cleared when the directory
  73 * data block containing the "." and ".." entries has been written.
  74 * MKDIR_PARENT is cleared when the parent inode with the increased link
  75 * count for ".." has been written. When both MKDIR flags have been
  76 * cleared, the DEPCOMPLETE flag is set to indicate that the directory
  77 * dependencies have been completed. The writing of the directory inode
  78 * itself sets the COMPLETE flag which then allows the directory entry for
  79 * the new directory to be written to disk. The RMDIR flag marks a dirrem
  80 * structure as representing the removal of a directory rather than a
  81 * file. When the removal dependencies are completed, additional work needs
  82 * to be done* (an additional decrement of the associated inode, and a
  83 * decrement of the parent inode).
  84 *
  85 * The DIRCHG flag marks a diradd structure as representing the changing
  86 * of an existing entry rather than the addition of a new one. When
  87 * the update is complete the dirrem associated with the inode for
  88 * the old name must be added to the worklist to do the necessary
  89 * reference count decrement.
  90 * 
  91 * The GOINGAWAY flag indicates that the data structure is frozen from
  92 * further change until its dependencies have been completed and its
  93 * resources freed after which it will be discarded.
  94 *
  95 * The IOSTARTED flag prevents multiple calls to the I/O start routine from
  96 * doing multiple rollbacks.
  97 *
  98 * The NEWBLOCK flag marks pagedep structures that have just been allocated,
  99 * so must be claimed by the inode before all dependencies are complete.
 100 *
 101 * The INPROGRESS flag marks worklist structures that are still on the
 102 * worklist, but are being considered for action by some process.
 103 *
 104 * The UFS1FMT flag indicates that the inode being processed is a ufs1 format.
 105 *
 106 * The EXTDATA flag indicates that the allocdirect describes an
 107 * extended-attributes dependency.
 108 *
 109 * The ONWORKLIST flag shows whether the structure is currently linked
 110 * onto a worklist.
 111 *
 112 * The UNLINK* flags track the progress of updating the on-disk linked
 113 * list of active but unlinked inodes. When an inode is first unlinked
 114 * it is marked as UNLINKED. When its on-disk di_freelink has been
 115 * written its UNLINKNEXT flags is set. When its predecessor in the
 116 * list has its di_freelink pointing at us its UNLINKPREV is set.
 117 * When the on-disk list can reach it from the superblock, its
 118 * UNLINKONLIST flag is set. Once all of these flags are set, it
 119 * is safe to let its last name be removed.
 120 */
 121#define	ATTACHED	0x000001
 122#define	UNDONE		0x000002
 123#define	COMPLETE	0x000004
 124#define	DEPCOMPLETE	0x000008
 125#define	MKDIR_PARENT	0x000010 /* diradd, mkdir, jaddref, jsegdep only */
 126#define	MKDIR_BODY	0x000020 /* diradd, mkdir, jaddref only */
 127#define	RMDIR		0x000040 /* dirrem only */
 128#define	DIRCHG		0x000080 /* diradd, dirrem only */
 129#define	GOINGAWAY	0x000100 /* indirdep, jremref only */
 130#define	IOSTARTED	0x000200 /* inodedep, pagedep, bmsafemap only */
 131#define	DELAYEDFREE	0x000400 /* allocindirect free delayed. */
 132#define	NEWBLOCK	0x000800 /* pagedep, jaddref only */
 133#define	INPROGRESS	0x001000 /* dirrem, freeblks, freefrag, freefile only */
 134#define	UFS1FMT		0x002000 /* indirdep only */
 135#define	EXTDATA		0x004000 /* allocdirect only */
 136#define	ONWORKLIST	0x008000
 137#define	IOWAITING	0x010000 /* Thread is waiting for IO to complete. */
 138#define	ONDEPLIST	0x020000 /* Structure is on a dependency list. */
 139#define	UNLINKED	0x040000 /* inodedep has been unlinked. */
 140#define	UNLINKNEXT	0x080000 /* inodedep has valid di_freelink */
 141#define	UNLINKPREV	0x100000 /* inodedep is pointed at in the unlink list */
 142#define	UNLINKONLIST	0x200000 /* inodedep is in the unlinked list on disk */
 143#define	UNLINKLINKS	(UNLINKNEXT | UNLINKPREV)
 144#define	WRITESUCCEEDED	0x400000 /* the disk write completed successfully */
 145
 146#define	ALLCOMPLETE	(ATTACHED | COMPLETE | DEPCOMPLETE)
 147
 148#define PRINT_SOFTDEP_FLAGS "\20\27writesucceeded\26unlinkonlist" \
 149	"\25unlinkprev\24unlinknext\23unlinked\22ondeplist\21iowaiting" \
 150	"\20onworklist\17extdata\16ufs1fmt\15inprogress\14newblock" \
 151	"\13delayedfree\12iostarted\11goingaway\10dirchg\7rmdir\6mkdir_body" \
 152	"\5mkdir_parent\4depcomplete\3complete\2undone\1attached"
 153
 154/*
 155 * Values for each of the soft dependency types.
 156 */
 157#define	D_UNUSED	0
 158#define	D_FIRST		D_PAGEDEP
 159#define	D_PAGEDEP	1
 160#define	D_INODEDEP	2
 161#define	D_BMSAFEMAP	3
 162#define	D_NEWBLK	4
 163#define	D_ALLOCDIRECT	5
 164#define	D_INDIRDEP	6
 165#define	D_ALLOCINDIR	7
 166#define	D_FREEFRAG	8
 167#define	D_FREEBLKS	9
 168#define	D_FREEFILE	10
 169#define	D_DIRADD	11
 170#define	D_MKDIR		12
 171#define	D_DIRREM	13
 172#define	D_NEWDIRBLK	14
 173#define	D_FREEWORK	15
 174#define	D_FREEDEP	16
 175#define	D_JADDREF	17
 176#define	D_JREMREF	18
 177#define	D_JMVREF	19
 178#define	D_JNEWBLK	20
 179#define	D_JFREEBLK	21
 180#define	D_JFREEFRAG	22
 181#define	D_JSEG		23
 182#define	D_JSEGDEP	24
 183#define	D_SBDEP		25
 184#define	D_JTRUNC	26
 185#define	D_JFSYNC	27
 186#define	D_SENTINEL	28
 187#define	D_LAST		D_SENTINEL
 188
 189/*
 190 * The workitem queue.
 191 * 
 192 * It is sometimes useful and/or necessary to clean up certain dependencies
 193 * in the background rather than during execution of an application process
 194 * or interrupt service routine. To realize this, we append dependency
 195 * structures corresponding to such tasks to a "workitem" queue. In a soft
 196 * updates implementation, most pending workitems should not wait for more
 197 * than a couple of seconds, so the filesystem syncer process awakens once
 198 * per second to process the items on the queue.
 199 */
 200
 201/* LIST_HEAD(workhead, worklist);	-- declared in buf.h */
 202
 203/*
 204 * Each request can be linked onto a work queue through its worklist structure.
 205 * To avoid the need for a pointer to the structure itself, this structure
 206 * MUST be declared FIRST in each type in which it appears! If more than one
 207 * worklist is needed in the structure, then a wk_data field must be added
 208 * and the macros below changed to use it.
 209 */
 210struct worklist {
 211	LIST_ENTRY(worklist)	wk_list;	/* list of work requests */
 212	struct mount		*wk_mp;		/* Mount we live in */
 213	unsigned int		wk_type:8,	/* type of request */
 214				wk_state:24;	/* state flags */
 215	LIST_ENTRY(worklist)	wk_all;		/* list of deps of this type */
 216#ifdef INVARIANTS
 217	const char		*wk_func;	/* func where added / removed */
 218	int			wk_line;	/* line where added / removed */
 219#endif
 220};
 221#define	WK_DATA(wk) ((void *)(wk))
 222#define	WK_PAGEDEP(wk) ((struct pagedep *)(wk))
 223#define	WK_INODEDEP(wk) ((struct inodedep *)(wk))
 224#define	WK_BMSAFEMAP(wk) ((struct bmsafemap *)(wk))
 225#define	WK_NEWBLK(wk)  ((struct newblk *)(wk))
 226#define	WK_ALLOCDIRECT(wk) ((struct allocdirect *)(wk))
 227#define	WK_INDIRDEP(wk) ((struct indirdep *)(wk))
 228#define	WK_ALLOCINDIR(wk) ((struct allocindir *)(wk))
 229#define	WK_FREEFRAG(wk) ((struct freefrag *)(wk))
 230#define	WK_FREEBLKS(wk) ((struct freeblks *)(wk))
 231#define	WK_FREEWORK(wk) ((struct freework *)(wk))
 232#define	WK_FREEFILE(wk) ((struct freefile *)(wk))
 233#define	WK_DIRADD(wk) ((struct diradd *)(wk))
 234#define	WK_MKDIR(wk) ((struct mkdir *)(wk))
 235#define	WK_DIRREM(wk) ((struct dirrem *)(wk))
 236#define	WK_NEWDIRBLK(wk) ((struct newdirblk *)(wk))
 237#define	WK_JADDREF(wk) ((struct jaddref *)(wk))
 238#define	WK_JREMREF(wk) ((struct jremref *)(wk))
 239#define	WK_JMVREF(wk) ((struct jmvref *)(wk))
 240#define	WK_JSEGDEP(wk) ((struct jsegdep *)(wk))
 241#define	WK_JSEG(wk) ((struct jseg *)(wk))
 242#define	WK_JNEWBLK(wk) ((struct jnewblk *)(wk))
 243#define	WK_JFREEBLK(wk) ((struct jfreeblk *)(wk))
 244#define	WK_FREEDEP(wk) ((struct freedep *)(wk))
 245#define	WK_JFREEFRAG(wk) ((struct jfreefrag *)(wk))
 246#define	WK_SBDEP(wk) ((struct sbdep *)(wk))
 247#define	WK_JTRUNC(wk) ((struct jtrunc *)(wk))
 248#define	WK_JFSYNC(wk) ((struct jfsync *)(wk))
 249
 250/*
 251 * Various types of lists
 252 */
 253LIST_HEAD(dirremhd, dirrem);
 254LIST_HEAD(diraddhd, diradd);
 255LIST_HEAD(newblkhd, newblk);
 256LIST_HEAD(inodedephd, inodedep);
 257LIST_HEAD(allocindirhd, allocindir);
 258LIST_HEAD(allocdirecthd, allocdirect);
 259TAILQ_HEAD(allocdirectlst, allocdirect);
 260LIST_HEAD(indirdephd, indirdep);
 261LIST_HEAD(jaddrefhd, jaddref);
 262LIST_HEAD(jremrefhd, jremref);
 263LIST_HEAD(jmvrefhd, jmvref);
 264LIST_HEAD(jnewblkhd, jnewblk);
 265LIST_HEAD(jblkdephd, jblkdep);
 266LIST_HEAD(freeworkhd, freework);
 267TAILQ_HEAD(freeworklst, freework);
 268TAILQ_HEAD(jseglst, jseg);
 269TAILQ_HEAD(inoreflst, inoref);
 270TAILQ_HEAD(freeblklst, freeblks);
 271
 272/*
 273 * The "pagedep" structure tracks the various dependencies related to
 274 * a particular directory page. If a directory page has any dependencies,
 275 * it will have a pagedep linked to its associated buffer. The
 276 * pd_dirremhd list holds the list of dirrem requests which decrement
 277 * inode reference counts. These requests are processed after the
 278 * directory page with the corresponding zero'ed entries has been
 279 * written. The pd_diraddhd list maintains the list of diradd requests
 280 * which cannot be committed until their corresponding inode has been
 281 * written to disk. Because a directory may have many new entries
 282 * being created, several lists are maintained hashed on bits of the
 283 * offset of the entry into the directory page to keep the lists from
 284 * getting too long. Once a new directory entry has been cleared to
 285 * be written, it is moved to the pd_pendinghd list. After the new
 286 * entry has been written to disk it is removed from the pd_pendinghd
 287 * list, any removed operations are done, and the dependency structure
 288 * is freed.
 289 */
 290#define	DAHASHSZ 5
 291#define	DIRADDHASH(offset) (((offset) >> 2) % DAHASHSZ)
 292struct pagedep {
 293	struct	worklist pd_list;	/* page buffer */
 294#	define	pd_state pd_list.wk_state /* check for multiple I/O starts */
 295	LIST_ENTRY(pagedep) pd_hash;	/* hashed lookup */
 296	ino_t	pd_ino;			/* associated file */
 297	ufs_lbn_t pd_lbn;		/* block within file */
 298	struct	newdirblk *pd_newdirblk; /* associated newdirblk if NEWBLOCK */
 299	struct	dirremhd pd_dirremhd;	/* dirrem's waiting for page */
 300	struct	diraddhd pd_diraddhd[DAHASHSZ]; /* diradd dir entry updates */
 301	struct	diraddhd pd_pendinghd;	/* directory entries awaiting write */
 302	struct	jmvrefhd pd_jmvrefhd;	/* Dependent journal writes. */
 303};
 304
 305/*
 306 * The "inodedep" structure tracks the set of dependencies associated
 307 * with an inode. One task that it must manage is delayed operations
 308 * (i.e., work requests that must be held until the inodedep's associated
 309 * inode has been written to disk). Getting an inode from its incore 
 310 * state to the disk requires two steps to be taken by the filesystem
 311 * in this order: first the inode must be copied to its disk buffer by
 312 * the VOP_UPDATE operation; second the inode's buffer must be written
 313 * to disk. To ensure that both operations have happened in the required
 314 * order, the inodedep maintains two lists. Delayed operations are
 315 * placed on the id_inowait list. When the VOP_UPDATE is done, all
 316 * operations on the id_inowait list are moved to the id_bufwait list.
 317 * When the buffer is written, the items on the id_bufwait list can be
 318 * safely moved to the work queue to be processed. A second task of the
 319 * inodedep structure is to track the status of block allocation within
 320 * the inode.  Each block that is allocated is represented by an
 321 * "allocdirect" structure (see below). It is linked onto the id_newinoupdt
 322 * list until both its contents and its allocation in the cylinder
 323 * group map have been written to disk. Once these dependencies have been
 324 * satisfied, it is removed from the id_newinoupdt list and any followup
 325 * actions such as releasing the previous block or fragment are placed
 326 * on the id_inowait list. When an inode is updated (a VOP_UPDATE is
 327 * done), the "inodedep" structure is linked onto the buffer through
 328 * its worklist. Thus, it will be notified when the buffer is about
 329 * to be written and when it is done. At the update time, all the
 330 * elements on the id_newinoupdt list are moved to the id_inoupdt list
 331 * since those changes are now relevant to the copy of the inode in the
 332 * buffer. Also at update time, the tasks on the id_inowait list are
 333 * moved to the id_bufwait list so that they will be executed when
 334 * the updated inode has been written to disk. When the buffer containing
 335 * the inode is written to disk, any updates listed on the id_inoupdt
 336 * list are rolled back as they are not yet safe. Following the write,
 337 * the changes are once again rolled forward and any actions on the
 338 * id_bufwait list are processed (since those actions are now safe).
 339 * The entries on the id_inoupdt and id_newinoupdt lists must be kept
 340 * sorted by logical block number to speed the calculation of the size
 341 * of the rolled back inode (see explanation in initiate_write_inodeblock).
 342 * When a directory entry is created, it is represented by a diradd.
 343 * The diradd is added to the id_inowait list as it cannot be safely
 344 * written to disk until the inode that it represents is on disk. After
 345 * the inode is written, the id_bufwait list is processed and the diradd
 346 * entries are moved to the id_pendinghd list where they remain until
 347 * the directory block containing the name has been written to disk.
 348 * The purpose of keeping the entries on the id_pendinghd list is so that
 349 * the softdep_fsync function can find and push the inode's directory
 350 * name(s) as part of the fsync operation for that file.
 351 */
 352struct inodedep {
 353	struct	worklist id_list;	/* buffer holding inode block */
 354#	define	id_state id_list.wk_state /* inode dependency state */
 355	LIST_ENTRY(inodedep) id_hash;	/* hashed lookup */
 356	TAILQ_ENTRY(inodedep) id_unlinked;	/* Unlinked but ref'd inodes */
 357	struct	fs *id_fs;		/* associated filesystem */
 358	ino_t	id_ino;			/* dependent inode */
 359	nlink_t	id_nlinkdelta;		/* saved effective link count */
 360	nlink_t	id_nlinkwrote;		/* i_nlink that we wrote to disk */
 361	nlink_t	id_savednlink;		/* Link saved during rollback */
 362	LIST_ENTRY(inodedep) id_deps;	/* bmsafemap's list of inodedep's */
 363	struct	bmsafemap *id_bmsafemap; /* related bmsafemap (if pending) */
 364	struct	diradd *id_mkdiradd;	/* diradd for a mkdir. */
 365	struct	inoreflst id_inoreflst;	/* Inode reference adjustments. */
 366	long	id_savedextsize;	/* ext size saved during rollback */
 367	off_t	id_savedsize;		/* file size saved during rollback */
 368	struct	dirremhd id_dirremhd;	/* Removals pending. */
 369	struct	workhead id_pendinghd;	/* entries awaiting directory write */
 370	struct	workhead id_bufwait;	/* operations after inode written */
 371	struct	workhead id_inowait;	/* operations waiting inode update */
 372	struct	allocdirectlst id_inoupdt; /* updates before inode written */
 373	struct	allocdirectlst id_newinoupdt; /* updates when inode written */
 374	struct	allocdirectlst id_extupdt; /* extdata updates pre-inode write */
 375	struct	allocdirectlst id_newextupdt; /* extdata updates at ino write */
 376	struct	freeblklst id_freeblklst; /* List of partial truncates. */
 377	union {
 378	struct	ufs1_dinode *idu_savedino1; /* saved ufs1_dinode contents */
 379	struct	ufs2_dinode *idu_savedino2; /* saved ufs2_dinode contents */
 380	} id_un;
 381};
 382#define	id_savedino1 id_un.idu_savedino1
 383#define	id_savedino2 id_un.idu_savedino2
 384
 385/*
 386 * A "bmsafemap" structure maintains a list of dependency structures
 387 * that depend on the update of a particular cylinder group map.
 388 * It has lists for newblks, allocdirects, allocindirs, and inodedeps.
 389 * It is attached to the buffer of a cylinder group block when any of
 390 * these things are allocated from the cylinder group. It is freed
 391 * after the cylinder group map is written and the state of its
 392 * dependencies are updated with DEPCOMPLETE to indicate that it has
 393 * been processed.
 394 */
 395struct bmsafemap {
 396	struct	worklist sm_list;	/* cylgrp buffer */
 397#	define	sm_state sm_list.wk_state
 398	LIST_ENTRY(bmsafemap) sm_hash;	/* Hash links. */
 399	LIST_ENTRY(bmsafemap) sm_next;	/* Mount list. */
 400	int	sm_cg;
 401	struct	buf *sm_buf;		/* associated buffer */
 402	struct	allocdirecthd sm_allocdirecthd; /* allocdirect deps */
 403	struct	allocdirecthd sm_allocdirectwr; /* writing allocdirect deps */
 404	struct	allocindirhd sm_allocindirhd; /* allocindir deps */
 405	struct	allocindirhd sm_allocindirwr; /* writing allocindir deps */
 406	struct	inodedephd sm_inodedephd; /* inodedep deps */
 407	struct	inodedephd sm_inodedepwr; /* writing inodedep deps */
 408	struct	newblkhd sm_newblkhd;	/* newblk deps */
 409	struct	newblkhd sm_newblkwr;	/* writing newblk deps */
 410	struct	jaddrefhd sm_jaddrefhd;	/* Pending inode allocations. */
 411	struct	jnewblkhd sm_jnewblkhd;	/* Pending block allocations. */
 412	struct	workhead sm_freehd;	/* Freedep deps. */
 413	struct	workhead sm_freewr;	/* Written freedeps. */
 414};
 415
 416/*
 417 * A "newblk" structure is attached to a bmsafemap structure when a block
 418 * or fragment is allocated from a cylinder group. Its state is set to
 419 * DEPCOMPLETE when its cylinder group map is written. It is converted to
 420 * an allocdirect or allocindir allocation once the allocator calls the
 421 * appropriate setup function. It will initially be linked onto a bmsafemap
 422 * list. Once converted it can be linked onto the lists described for
 423 * allocdirect or allocindir as described below.
 424 */ 
 425struct newblk {
 426	struct	worklist nb_list;	/* See comment above. */
 427#	define	nb_state nb_list.wk_state
 428	LIST_ENTRY(newblk) nb_hash;	/* Hashed lookup. */
 429	LIST_ENTRY(newblk) nb_deps;	/* Bmsafemap's list of newblks. */
 430	struct	jnewblk *nb_jnewblk;	/* New block journal entry. */
 431	struct	bmsafemap *nb_bmsafemap;/* Cylgrp dep (if pending). */
 432	struct	freefrag *nb_freefrag;	/* Fragment to be freed (if any). */
 433	struct	indirdephd nb_indirdeps; /* Children indirect blocks. */
 434	struct	workhead nb_newdirblk;	/* Dir block to notify when written. */
 435	struct	workhead nb_jwork;	/* Journal work pending. */
 436	ufs2_daddr_t	nb_newblkno;	/* New value of block pointer. */
 437};
 438
 439/*
 440 * An "allocdirect" structure is attached to an "inodedep" when a new block
 441 * or fragment is allocated and pointed to by the inode described by
 442 * "inodedep". The worklist is linked to the buffer that holds the block.
 443 * When the block is first allocated, it is linked to the bmsafemap
 444 * structure associated with the buffer holding the cylinder group map
 445 * from which it was allocated. When the cylinder group map is written
 446 * to disk, ad_state has the DEPCOMPLETE flag set. When the block itself
 447 * is written, the COMPLETE flag is set. Once both the cylinder group map
 448 * and the data itself have been written, it is safe to write the inode
 449 * that claims the block. If there was a previous fragment that had been
 450 * allocated before the file was increased in size, the old fragment may
 451 * be freed once the inode claiming the new block is written to disk.
 452 * This ad_fragfree request is attached to the id_inowait list of the
 453 * associated inodedep (pointed to by ad_inodedep) for processing after
 454 * the inode is written. When a block is allocated to a directory, an
 455 * fsync of a file whose name is within that block must ensure not only
 456 * that the block containing the file name has been written, but also
 457 * that the on-disk inode references that block. When a new directory
 458 * block is created, we allocate a newdirblk structure which is linked
 459 * to the associated allocdirect (on its ad_newdirblk list). When the
 460 * allocdirect has been satisfied, the newdirblk structure is moved to
 461 * the inodedep id_bufwait list of its directory to await the inode
 462 * being written. When the inode is written, the directory entries are
 463 * fully committed and can be deleted from their pagedep->id_pendinghd
 464 * and inodedep->id_pendinghd lists.
 465 */
 466struct allocdirect {
 467	struct	newblk ad_block;	/* Common block logic */
 468#	define	ad_list ad_block.nb_list /* block pointer worklist */
 469#	define	ad_state ad_list.wk_state /* block pointer state */
 470	TAILQ_ENTRY(allocdirect) ad_next; /* inodedep's list of allocdirect's */
 471	struct	inodedep *ad_inodedep;	/* associated inodedep */
 472	ufs2_daddr_t	ad_oldblkno;	/* old value of block pointer */
 473	int		ad_offset;	/* Pointer offset in parent. */
 474	long		ad_newsize;	/* size of new block */
 475	long		ad_oldsize;	/* size of old block */
 476};
 477#define	ad_newblkno	ad_block.nb_newblkno
 478#define	ad_freefrag	ad_block.nb_freefrag
 479#define	ad_newdirblk	ad_block.nb_newdirblk
 480
 481/*
 482 * A single "indirdep" structure manages all allocation dependencies for
 483 * pointers in an indirect block. The up-to-date state of the indirect
 484 * block is stored in ir_savedata. The set of pointers that may be safely
 485 * written to the disk is stored in ir_savebp. The state field is used
 486 * only to track whether the buffer is currently being written (in which
 487 * case it is not safe to update ir_savebp). Ir_deplisthd contains the
 488 * list of allocindir structures, one for each block that needs to be
 489 * written to disk. Once the block and its bitmap allocation have been
 490 * written the safecopy can be updated to reflect the allocation and the
 491 * allocindir structure freed. If ir_state indicates that an I/O on the
 492 * indirect block is in progress when ir_savebp is to be updated, the
 493 * update is deferred by placing the allocindir on the ir_donehd list.
 494 * When the I/O on the indirect block completes, the entries on the
 495 * ir_donehd list are processed by updating their corresponding ir_savebp
 496 * pointers and then freeing the allocindir structure.
 497 */
 498struct indirdep {
 499	struct	worklist ir_list;	/* buffer holding indirect block */
 500#	define	ir_state ir_list.wk_state /* indirect block pointer state */
 501	LIST_ENTRY(indirdep) ir_next;	/* alloc{direct,indir} list */
 502	TAILQ_HEAD(, freework) ir_trunc;	/* List of truncations. */
 503	caddr_t	ir_saveddata;		/* buffer cache contents */
 504	struct	buf *ir_savebp;		/* buffer holding safe copy */
 505	struct	buf *ir_bp;		/* buffer holding live copy */
 506	struct	allocindirhd ir_completehd; /* waiting for indirdep complete */
 507	struct	allocindirhd ir_writehd; /* Waiting for the pointer write. */
 508	struct	allocindirhd ir_donehd;	/* done waiting to update safecopy */
 509	struct	allocindirhd ir_deplisthd; /* allocindir deps for this block */
 510	struct	freeblks *ir_freeblks;	/* Freeblks that frees this indir. */
 511};
 512
 513/*
 514 * An "allocindir" structure is attached to an "indirdep" when a new block
 515 * is allocated and pointed to by the indirect block described by the
 516 * "indirdep". The worklist is linked to the buffer that holds the new block.
 517 * When the block is first allocated, it is linked to the bmsafemap
 518 * structure associated with the buffer holding the cylinder group map
 519 * from which it was allocated. When the cylinder group map is written
 520 * to disk, ai_state has the DEPCOMPLETE flag set. When the block itself
 521 * is written, the COMPLETE flag is set. Once both the cylinder group map
 522 * and the data itself have been written, it is safe to write the entry in
 523 * the indirect block that claims the block; the "allocindir" dependency 
 524 * can then be freed as it is no longer applicable.
 525 */
 526struct allocindir {
 527	struct	newblk ai_block;	/* Common block area */
 528#	define	ai_state ai_block.nb_list.wk_state /* indirect pointer state */
 529	LIST_ENTRY(allocindir) ai_next;	/* indirdep's list of allocindir's */
 530	struct	indirdep *ai_indirdep;	/* address of associated indirdep */
 531	ufs2_daddr_t	ai_oldblkno;	/* old value of block pointer */
 532	ufs_lbn_t	ai_lbn;		/* Logical block number. */
 533	int		ai_offset;	/* Pointer offset in parent. */
 534};
 535#define	ai_newblkno	ai_block.nb_newblkno
 536#define	ai_freefrag	ai_block.nb_freefrag
 537#define	ai_newdirblk	ai_block.nb_newdirblk
 538
 539/*
 540 * The allblk union is used to size the newblk structure on allocation so
 541 * that it may be any one of three types.
 542 */
 543union allblk {
 544	struct	allocindir ab_allocindir;
 545	struct	allocdirect ab_allocdirect;
 546	struct	newblk	ab_newblk;
 547};
 548
 549/*
 550 * A "freefrag" structure is attached to an "inodedep" when a previously
 551 * allocated fragment is replaced with a larger fragment, rather than extended.
 552 * The "freefrag" structure is constructed and attached when the replacement
 553 * block is first allocated. It is processed after the inode claiming the
 554 * bigger block that replaces it has been written to disk.
 555 */
 556struct freefrag {
 557	struct	worklist ff_list;	/* id_inowait or delayed worklist */
 558#	define	ff_state ff_list.wk_state
 559	struct	worklist *ff_jdep;	/* Associated journal entry. */
 560	struct	workhead ff_jwork;	/* Journal work pending. */
 561	ufs2_daddr_t ff_blkno;		/* fragment physical block number */
 562	long	ff_fragsize;		/* size of fragment being deleted */
 563	ino_t	ff_inum;		/* owning inode number */
 564	__enum_uint8(vtype) ff_vtype;	/* owning inode's file type */
 565	int	ff_key;			/* trim key when deleted */
 566};
 567
 568/*
 569 * A "freeblks" structure is attached to an "inodedep" when the
 570 * corresponding file's length is reduced to zero. It records all
 571 * the information needed to free the blocks of a file after its
 572 * zero'ed inode has been written to disk.  The actual work is done
 573 * by child freework structures which are responsible for individual
 574 * inode pointers while freeblks is responsible for retiring the
 575 * entire operation when it is complete and holding common members.
 576 */
 577struct freeblks {
 578	struct	worklist fb_list;	/* id_inowait or delayed worklist */
 579#	define	fb_state fb_list.wk_state /* inode and dirty block state */
 580	TAILQ_ENTRY(freeblks) fb_next;	/* List of inode truncates. */
 581	struct	jblkdephd fb_jblkdephd;	/* Journal entries pending */
 582	struct	workhead fb_freeworkhd;	/* Work items pending */
 583	struct	workhead fb_jwork;	/* Journal work pending */
 584	struct	vnode *fb_devvp;	/* filesystem device vnode */
 585#ifdef QUOTA
 586	struct	dquot *fb_quota[MAXQUOTAS]; /* quotas to be adjusted */
 587#endif
 588	uint64_t fb_modrev;		/* Inode revision at start of trunc. */
 589	off_t	fb_len;			/* Length we're truncating to. */
 590	ufs2_daddr_t fb_chkcnt;		/* Blocks released. */
 591	ino_t	fb_inum;		/* inode owner of blocks */
 592	__enum_uint8(vtype) fb_vtype;	/* inode owner's file type */
 593	uid_t	fb_uid;			/* uid of previous owner of blocks */
 594	int	fb_ref;			/* Children outstanding. */
 595	int	fb_cgwait;		/* cg writes outstanding. */
 596};
 597
 598/*
 599 * A "freework" structure handles the release of a tree of blocks or a single
 600 * block.  Each indirect block in a tree is allocated its own freework
 601 * structure so that the indirect block may be freed only when all of its
 602 * children are freed.  In this way we enforce the rule that an allocated
 603 * block must have a valid path to a root that is journaled.  Each child
 604 * block acquires a reference and when the ref hits zero the parent ref
 605 * is decremented.  If there is no parent the freeblks ref is decremented.
 606 */
 607struct freework {
 608	struct	worklist fw_list;		/* Delayed worklist. */
 609#	define	fw_state fw_list.wk_state
 610	LIST_ENTRY(freework) fw_segs;		/* Seg list. */
 611	TAILQ_ENTRY(freework) fw_next;		/* Hash/Trunc list. */
 612	struct	jnewblk	 *fw_jnewblk;		/* Journal entry to cancel. */
 613	struct	freeblks *fw_freeblks;		/* Root of operation. */
 614	struct	freework *fw_parent;		/* Parent indirect. */
 615	struct	indirdep *fw_indir;		/* indirect block. */
 616	ufs2_daddr_t	 fw_blkno;		/* Our block #. */
 617	ufs_lbn_t	 fw_lbn;		/* Original lbn before free. */
 618	uint16_t	 fw_frags;		/* Number of frags. */
 619	uint16_t	 fw_ref;		/* Number of children out. */
 620	uint16_t	 fw_off;		/* Current working position. */
 621	uint16_t	 fw_start;		/* Start of partial truncate. */
 622};
 623
 624/*
 625 * A "freedep" structure is allocated to track the completion of a bitmap
 626 * write for a freework.  One freedep may cover many freed blocks so long
 627 * as they reside in the same cylinder group.  When the cg is written
 628 * the freedep decrements the ref on the freework which may permit it
 629 * to be freed as well.
 630 */
 631struct freedep {
 632	struct	worklist fd_list;	/* Delayed worklist. */
 633	struct	freework *fd_freework;	/* Parent freework. */
 634};
 635
 636/*
 637 * A "freefile" structure is attached to an inode when its
 638 * link count is reduced to zero. It marks the inode as free in
 639 * the cylinder group map after the zero'ed inode has been written
 640 * to disk and any associated blocks and fragments have been freed.
 641 */
 642struct freefile {
 643	struct	worklist fx_list;	/* id_inowait or delayed worklist */
 644	mode_t	fx_mode;		/* mode of inode */
 645	ino_t	fx_oldinum;		/* inum of the unlinked file */
 646	struct	vnode *fx_devvp;	/* filesystem device vnode */
 647	struct	workhead fx_jwork;	/* journal work pending. */
 648};
 649
 650/*
 651 * A "diradd" structure is linked to an "inodedep" id_inowait list when a
 652 * new directory entry is allocated that references the inode described
 653 * by "inodedep". When the inode itself is written (either the initial
 654 * allocation for new inodes or with the increased link count for
 655 * existing inodes), the COMPLETE flag is set in da_state. If the entry
 656 * is for a newly allocated inode, the "inodedep" structure is associated
 657 * with a bmsafemap which prevents the inode from being written to disk
 658 * until the cylinder group has been updated. Thus the da_state COMPLETE
 659 * flag cannot be set until the inode bitmap dependency has been removed.
 660 * When creating a new file, it is safe to write the directory entry that
 661 * claims the inode once the referenced inode has been written. Since
 662 * writing the inode clears the bitmap dependencies, the DEPCOMPLETE flag
 663 * in the diradd can be set unconditionally when creating a file. When
 664 * creating a directory, there are two additional dependencies described by
 665 * mkdir structures (see their description below). When these dependencies
 666 * are resolved the DEPCOMPLETE flag is set in the diradd structure.
 667 * If there are multiple links created to the same inode, there will be
 668 * a separate diradd structure created for each link. The diradd is
 669 * linked onto the pg_diraddhd list of the pagedep for the directory
 670 * page that contains the entry. When a directory page is written,
 671 * the pg_diraddhd list is traversed to rollback any entries that are
 672 * not yet ready to be written to disk. If a directory entry is being
 673 * changed (by rename) rather than added, the DIRCHG flag is set and
 674 * the da_previous entry points to the entry that will be "removed"
 675 * once the new entry has been committed. During rollback, entries
 676 * with da_previous are replaced with the previous inode number rather
 677 * than zero.
 678 *
 679 * The overlaying of da_pagedep and da_previous is done to keep the
 680 * structure down. If a da_previous entry is present, the pointer to its
 681 * pagedep is available in the associated dirrem entry. If the DIRCHG flag
 682 * is set, the da_previous entry is valid; if not set the da_pagedep entry
 683 * is valid. The DIRCHG flag never changes; it is set when the structure
 684 * is created if appropriate and is never cleared.
 685 */
 686struct diradd {
 687	struct	worklist da_list;	/* id_inowait or id_pendinghd list */
 688#	define	da_state da_list.wk_state /* state of the new directory entry */
 689	LIST_ENTRY(diradd) da_pdlist;	/* pagedep holding directory block */
 690	doff_t	da_offset;		/* offset of new dir entry in dir blk */
 691	ino_t	da_newinum;		/* inode number for the new dir entry */
 692	union {
 693	struct	dirrem *dau_previous;	/* entry being replaced in dir change */
 694	struct	pagedep *dau_pagedep;	/* pagedep dependency for addition */
 695	} da_un;
 696	struct workhead da_jwork;	/* Journal work awaiting completion. */
 697};
 698#define	da_previous da_un.dau_previous
 699#define	da_pagedep da_un.dau_pagedep
 700
 701/*
 702 * Two "mkdir" structures are needed to track the additional dependencies
 703 * associated with creating a new directory entry. Normally a directory
 704 * addition can be committed as soon as the newly referenced inode has been
 705 * written to disk with its increased link count. When a directory is
 706 * created there are two additional dependencies: writing the directory
 707 * data block containing the "." and ".." entries (MKDIR_BODY) and writing
 708 * the parent inode with the increased link count for ".." (MKDIR_PARENT).
 709 * These additional dependencies are tracked by two mkdir structures that
 710 * reference the associated "diradd" structure. When they have completed,
 711 * they set the DEPCOMPLETE flag on the diradd so that it knows that its
 712 * extra dependencies have been completed. The md_state field is used only
 713 * to identify which type of dependency the mkdir structure is tracking.
 714 * It is not used in the mainline code for any purpose other than consistency
 715 * checking. All the mkdir structures in the system are linked together on
 716 * a list. This list is needed so that a diradd can find its associated
 717 * mkdir structures and deallocate them if it is prematurely freed (as for
 718 * example if a mkdir is immediately followed by a rmdir of the same directory).
 719 * Here, the free of the diradd must traverse the list to find the associated
 720 * mkdir structures that reference it. The deletion would be faster if the
 721 * diradd structure were simply augmented to have two pointers that referenced
 722 * the associated mkdir's. However, this would increase the size of the diradd
 723 * structure to speed a very infrequent operation.
 724 */
 725struct mkdir {
 726	struct	worklist md_list;	/* id_inowait or buffer holding dir */
 727#	define	md_state md_list.wk_state /* type: MKDIR_PARENT or MKDIR_BODY */
 728	struct	diradd *md_diradd;	/* associated diradd */
 729	struct	jaddref *md_jaddref;	/* dependent jaddref. */
 730	struct	buf *md_buf;		/* MKDIR_BODY: buffer holding dir */
 731	LIST_ENTRY(mkdir) md_mkdirs;	/* list of all mkdirs */
 732};
 733
 734/*
 735 * A "dirrem" structure describes an operation to decrement the link
 736 * count on an inode. The dirrem structure is attached to the pg_dirremhd
 737 * list of the pagedep for the directory page that contains the entry.
 738 * It is processed after the directory page with the deleted entry has
 739 * been written to disk.
 740 */
 741struct dirrem {
 742	struct	worklist dm_list;	/* delayed worklist */
 743#	define	dm_state dm_list.wk_state /* state of the old directory entry */
 744	LIST_ENTRY(dirrem) dm_next;	/* pagedep's list of dirrem's */
 745	LIST_ENTRY(dirrem) dm_inonext;	/* inodedep's list of dirrem's */
 746	struct	jremrefhd dm_jremrefhd;	/* Pending remove reference deps. */
 747	ino_t	dm_oldinum;		/* inum of the removed dir entry */
 748	doff_t	dm_offset;		/* offset of removed dir entry in blk */
 749	union {
 750	struct	pagedep *dmu_pagedep;	/* pagedep dependency for remove */
 751	ino_t	dmu_dirinum;		/* parent inode number (for rmdir) */
 752	} dm_un;
 753	struct workhead dm_jwork;	/* Journal work awaiting completion. */
 754};
 755#define	dm_pagedep dm_un.dmu_pagedep
 756#define	dm_dirinum dm_un.dmu_dirinum
 757
 758/*
 759 * A "newdirblk" structure tracks the progress of a newly allocated
 760 * directory block from its creation until it is claimed by its on-disk
 761 * inode. When a block is allocated to a directory, an fsync of a file
 762 * whose name is within that block must ensure not only that the block
 763 * containing the file name has been written, but also that the on-disk
 764 * inode references that block. When a new directory block is created,
 765 * we allocate a newdirblk structure which is linked to the associated
 766 * allocdirect (on its ad_newdirblk list). When the allocdirect has been
 767 * satisfied, the newdirblk structure is moved to the inodedep id_bufwait
 768 * list of its directory to await the inode being written. When the inode
 769 * is written, the directory entries are fully committed and can be
 770 * deleted from their pagedep->id_pendinghd and inodedep->id_pendinghd
 771 * lists. Note that we could track directory blocks allocated to indirect
 772 * blocks using a similar scheme with the allocindir structures. Rather
 773 * than adding this level of complexity, we simply write those newly 
 774 * allocated indirect blocks synchronously as such allocations are rare.
 775 * In the case of a new directory the . and .. links are tracked with
 776 * a mkdir rather than a pagedep.  In this case we track the mkdir
 777 * so it can be released when it is written.  A workhead is used
 778 * to simplify canceling a mkdir that is removed by a subsequent dirrem.
 779 */
 780struct newdirblk {
 781	struct	worklist db_list;	/* id_inowait or pg_newdirblk */
 782#	define	db_state db_list.wk_state
 783	struct	pagedep *db_pagedep;	/* associated pagedep */
 784	struct	workhead db_mkdir;
 785};
 786
 787/*
 788 * The inoref structure holds the elements common to jaddref and jremref
 789 * so they may easily be queued in-order on the inodedep.
 790 */
 791struct inoref {
 792	struct	worklist if_list;	/* Journal pending or jseg entries. */
 793#	define	if_state if_list.wk_state
 794	TAILQ_ENTRY(inoref) if_deps;	/* Links for inodedep. */
 795	struct	jsegdep	*if_jsegdep;	/* Will track our journal record. */
 796	off_t		if_diroff;	/* Directory offset. */
 797	ino_t		if_ino;		/* Inode number. */
 798	ino_t		if_parent;	/* Parent inode number. */
 799	nlink_t		if_nlink;	/* nlink before addition. */
 800	uint16_t	if_mode;	/* File mode, needed for IFMT. */
 801};
 802
 803/*
 804 * A "jaddref" structure tracks a new reference (link count) on an inode
 805 * and prevents the link count increase and bitmap allocation until a
 806 * journal entry can be written.  Once the journal entry is written,
 807 * the inode is put on the pendinghd of the bmsafemap and a diradd or
 808 * mkdir entry is placed on the bufwait list of the inode.  The DEPCOMPLETE
 809 * flag is used to indicate that all of the required information for writing
 810 * the journal entry is present.  MKDIR_BODY and MKDIR_PARENT are used to
 811 * differentiate . and .. links from regular file names.  NEWBLOCK indicates
 812 * a bitmap is still pending.  If a new reference is canceled by a delete
 813 * prior to writing the journal the jaddref write is canceled and the
 814 * structure persists to prevent any disk-visible changes until it is
 815 * ultimately released when the file is freed or the link is dropped again.
 816 */
 817struct jaddref {
 818	struct	inoref	ja_ref;		/* see inoref above. */
 819#	define	ja_list	ja_ref.if_list	/* Jrnl pending, id_inowait, dm_jwork.*/
 820#	define	ja_state ja_ref.if_list.wk_state
 821	LIST_ENTRY(jaddref) ja_bmdeps;	/* Links for bmsafemap. */
 822	union {
 823		struct	diradd	*jau_diradd;	/* Pending diradd. */
 824		struct	mkdir	*jau_mkdir;	/* MKDIR_{PARENT,BODY} */
 825	} ja_un;
 826};
 827#define	ja_diradd	ja_un.jau_diradd
 828#define	ja_mkdir	ja_un.jau_mkdir
 829#define	ja_diroff	ja_ref.if_diroff
 830#define	ja_ino		ja_ref.if_ino
 831#define	ja_parent	ja_ref.if_parent
 832#define	ja_mode		ja_ref.if_mode
 833
 834/*
 835 * A "jremref" structure tracks a removed reference (unlink) on an
 836 * inode and prevents the directory remove from proceeding until the
 837 * journal entry is written.  Once the journal has been written the remove
 838 * may proceed as normal. 
 839 */
 840struct jremref {
 841	struct	inoref	jr_ref;		/* see inoref above. */
 842#	define	jr_list	jr_ref.if_list	/* Linked to softdep_journal_pending. */
 843#	define	jr_state jr_ref.if_list.wk_state
 844	LIST_ENTRY(jremref) jr_deps;	/* Links for dirrem. */
 845	struct	dirrem	*jr_dirrem;	/* Back pointer to dirrem. */
 846};
 847
 848/*
 849 * A "jmvref" structure tracks a name relocations within the same
 850 * directory block that occur as a result of directory compaction.
 851 * It prevents the updated directory entry from being written to disk
 852 * until the journal entry is written. Once the journal has been
 853 * written the compacted directory may be written to disk.
 854 */
 855struct jmvref {
 856	struct	worklist jm_list;	/* Linked to softdep_journal_pending. */
 857	LIST_ENTRY(jmvref) jm_deps;	/* Jmvref on pagedep. */
 858	struct pagedep	*jm_pagedep;	/* Back pointer to pagedep. */
 859	ino_t		jm_parent;	/* Containing directory inode number. */
 860	ino_t		jm_ino;		/* Inode number of our entry. */
 861	off_t		jm_oldoff;	/* Our old offset in directory. */
 862	off_t		jm_newoff;	/* Our new offset in directory. */
 863};
 864
 865/*
 866 * A "jnewblk" structure tracks a newly allocated block or fragment and
 867 * prevents the direct or indirect block pointer as well as the cg bitmap
 868 * from being written until it is logged.  After it is logged the jsegdep
 869 * is attached to the allocdirect or allocindir until the operation is
 870 * completed or reverted.  If the operation is reverted prior to the journal
 871 * write the jnewblk structure is maintained to prevent the bitmaps from
 872 * reaching the disk.  Ultimately the jnewblk structure will be passed
 873 * to the free routine as the in memory cg is modified back to the free
 874 * state at which time it can be released. It may be held on any of the
 875 * fx_jwork, fw_jwork, fb_jwork, ff_jwork, nb_jwork, or ir_jwork lists.
 876 */
 877struct jnewblk {
 878	struct	worklist jn_list;	/* See lists above. */
 879#	define	jn_state jn_list.wk_state
 880	struct	jsegdep	*jn_jsegdep;	/* Will track our journal record. */
 881	LIST_ENTRY(jnewblk) jn_deps;	/* Jnewblks on sm_jnewblkhd. */
 882	struct	worklist *jn_dep;	/* Dependency to ref completed seg. */
 883	ufs_lbn_t	jn_lbn;		/* Lbn to which allocated. */
 884	ufs2_daddr_t	jn_blkno;	/* Blkno allocated */
 885	ino_t		jn_ino;		/* Ino to which allocated. */
 886	int		jn_oldfrags;	/* Previous fragments when extended. */
 887	int		jn_frags;	/* Number of fragments. */
 888};
 889
 890/*
 891 * A "jblkdep" structure tracks jfreeblk and jtrunc records attached to a
 892 * freeblks structure.
 893 */
 894struct jblkdep {
 895	struct	worklist jb_list;	/* For softdep journal pending. */
 896	struct	jsegdep *jb_jsegdep;	/* Reference to the jseg. */
 897	struct	freeblks *jb_freeblks;	/* Back pointer to freeblks. */
 898	LIST_ENTRY(jblkdep) jb_deps;	/* Dep list on freeblks. */
 899
 900};
 901
 902/*
 903 * A "jfreeblk" structure tracks the journal write for freeing a block
 904 * or tree of blocks.  The block pointer must not be cleared in the inode
 905 * or indirect prior to the jfreeblk being written to the journal.
 906 */
 907struct jfreeblk {
 908	struct	jblkdep	jf_dep;		/* freeblks linkage. */
 909	ufs_lbn_t	jf_lbn;		/* Lbn from which blocks freed. */
 910	ufs2_daddr_t	jf_blkno;	/* Blkno being freed. */
 911	ino_t		jf_ino;		/* Ino from which blocks freed. */
 912	int		jf_frags;	/* Number of frags being freed. */
 913};
 914
 915/*
 916 * A "jfreefrag" tracks the freeing of a single block when a fragment is
 917 * extended or an indirect page is replaced.  It is not part of a larger
 918 * freeblks operation.
 919 */
 920struct jfreefrag {
 921	struct	worklist fr_list;	/* Linked to softdep_journal_pending. */
 922#	define	fr_state fr_list.wk_state
 923	struct	jsegdep	*fr_jsegdep;	/* Will track our journal record. */
 924	struct freefrag	*fr_freefrag;	/* Back pointer to freefrag. */
 925	ufs_lbn_t	fr_lbn;		/* Lbn from which frag freed. */
 926	ufs2_daddr_t	fr_blkno;	/* Blkno being freed. */
 927	ino_t		fr_ino;		/* Ino from which frag freed. */
 928	int		fr_frags;	/* Size of frag being freed. */
 929};
 930
 931/*
 932 * A "jtrunc" journals the intent to truncate an inode's data or extent area.
 933 */
 934struct jtrunc {
 935	struct	jblkdep	jt_dep;		/* freeblks linkage. */
 936	off_t		jt_size;	/* Final file size. */
 937	int		jt_extsize;	/* Final extent size. */
 938	ino_t		jt_ino;		/* Ino being truncated. */
 939};
 940
 941/*
 942 * A "jfsync" journals the completion of an fsync which invalidates earlier
 943 * jtrunc records in the journal.
 944 */
 945struct jfsync {
 946	struct worklist	jfs_list;	/* For softdep journal pending. */
 947	off_t		jfs_size;	/* Sync file size. */
 948	int		jfs_extsize;	/* Sync extent size. */
 949	ino_t		jfs_ino;	/* ino being synced. */
 950};
 951
 952/*
 953 * A "jsegdep" structure tracks a single reference to a written journal
 954 * segment so the journal space can be reclaimed when all dependencies
 955 * have been written. It can hang off of id_inowait, dm_jwork, da_jwork,
 956 * nb_jwork, ff_jwork, or fb_jwork lists.
 957 */
 958struct jsegdep {
 959	struct	worklist jd_list;	/* See above for lists. */
 960#	define	jd_state jd_list.wk_state
 961	struct	jseg	*jd_seg;	/* Our journal record. */
 962};
 963
 964/*
 965 * A "jseg" structure contains all of the journal records written in a
 966 * single disk write.  The jaddref and jremref structures are linked into
 967 * js_entries so thay may be completed when the write completes.  The
 968 * js_entries also include the write dependency structures: jmvref,
 969 * jnewblk, jfreeblk, jfreefrag, and jtrunc.  The js_refs field counts
 970 * the number of entries on the js_entries list. Thus there is a single
 971 * jseg entry to describe each journal write.
 972 */
 973struct jseg {
 974	struct	worklist js_list;	/* b_deps link for journal */
 975#	define	js_state js_list.wk_state
 976	struct	workhead js_entries;	/* Entries awaiting write */
 977	LIST_HEAD(, freework) js_indirs;/* List of indirects in this seg. */
 978	TAILQ_ENTRY(jseg) js_next;	/* List of all unfinished segments. */
 979	struct	jblocks *js_jblocks;	/* Back pointer to block/seg list */
 980	struct	buf *js_buf;		/* Buffer while unwritten */
 981	uint64_t js_seq;		/* Journal record sequence number. */
 982	uint64_t js_oldseq;		/* Oldest valid sequence number. */
 983	int	js_size;		/* Size of journal record in bytes. */
 984	int	js_cnt;			/* Total items allocated. */
 985	int	js_refs;		/* Count of js_entries items. */
 986};
 987
 988/*
 989 * A 'sbdep' structure tracks the head of the free inode list and
 990 * superblock writes.  This makes sure the superblock is always pointing at
 991 * the first possible unlinked inode for the suj recovery process.  If a
 992 * block write completes and we discover a new head is available the buf
 993 * is dirtied and the dep is kept. See the description of the UNLINK*
 994 * flags above for more details.
 995 */
 996struct sbdep {
 997	struct	worklist sb_list;	/* b_dep linkage */
 998	struct	fs	*sb_fs;		/* Filesystem pointer within buf. */
 999	struct	ufsmount *sb_ump;	/* Our mount structure */
1000};
1001
1002/*
1003 * Private journaling structures.
1004 */
1005struct jblocks {
1006	struct jseglst	jb_segs;	/* TAILQ of current segments. */
1007	struct jseg	*jb_writeseg;	/* Next write to complete. */
1008	struct jseg	*jb_oldestseg;	/* Oldest segment with valid entries. */
1009	struct jextent	*jb_extent;	/* Extent array. */
1010	uint64_t	jb_nextseq;	/* Next sequence number. */
1011	uint64_t	jb_oldestwrseq;	/* Oldest written sequence number. */
1012	uint8_t		jb_needseg;	/* Need a forced segment. */
1013	uint8_t		jb_suspended;	/* Did journal suspend writes? */
1014	int		jb_avail;	/* Available extents. */
1015	int		jb_used;	/* Last used extent. */
1016	int		jb_head;	/* Allocator head. */
1017	int		jb_off;		/* Allocator extent offset. */
1018	int		jb_blocks;	/* Total disk blocks covered. */
1019	int		jb_free;	/* Total disk blocks free. */
1020	int		jb_min;		/* Minimum free space. */
1021	int		jb_low;		/* Low on space. */
1022	int		jb_age;		/* Insertion time of oldest rec. */
1023};
1024
1025struct jextent {
1026	ufs2_daddr_t	je_daddr;	/* Disk block address. */
1027	int		je_blocks;	/* Disk block count. */
1028};
1029
1030/*
1031 * Hash table declarations.
1032 */
1033LIST_HEAD(mkdirlist, mkdir);
1034LIST_HEAD(pagedep_hashhead, pagedep);
1035LIST_HEAD(inodedep_hashhead, inodedep);
1036LIST_HEAD(newblk_hashhead, newblk);
1037LIST_HEAD(bmsafemap_hashhead, bmsafemap);
1038TAILQ_HEAD(indir_hashhead, freework);
1039
1040/*
1041 * Per-filesystem soft dependency data.
1042 * Allocated at mount and freed at unmount.
1043 */
1044struct mount_softdeps {
1045	struct	rwlock sd_fslock;		/* softdep lock */
1046	struct	workhead sd_workitem_pending;	/* softdep work queue */
1047	struct	worklist *sd_worklist_tail;	/* Tail pointer for above */
1048	struct	workhead sd_journal_pending;	/* journal work queue */
1049	struct	worklist *sd_journal_tail;	/* Tail pointer for above */
1050	struct	jblocks *sd_jblocks;		/* Journal block information */
1051	struct	inodedeplst sd_unlinked;	/* Unlinked inodes */
1052	struct	bmsafemaphd sd_dirtycg;		/* Dirty CGs */
1053	struct	mkdirlist sd_mkdirlisthd;	/* Track mkdirs */
1054	struct	pagedep_hashhead *sd_pdhash;	/* pagedep hash table */
1055	u_long	sd_pdhashsize;			/* pagedep hash table size-1 */
1056	long	sd_pdnextclean;			/* next hash bucket to clean */
1057	struct	inodedep_hashhead *sd_idhash;	/* inodedep hash table */
1058	u_long	sd_idhashsize;			/* inodedep hash table size-1 */
1059	long	sd_idnextclean;			/* next hash bucket to clean */
1060	struct	newblk_hashhead *sd_newblkhash;	/* newblk hash table */
1061	u_long	sd_newblkhashsize;		/* newblk hash table size-1 */
1062	struct	bmsafemap_hashhead *sd_bmhash;	/* bmsafemap hash table */
1063	u_long	sd_bmhashsize;			/* bmsafemap hash table size-1*/
1064	struct	indir_hashhead *sd_indirhash;	/* indir hash table */
1065	uint64_t sd_indirhashsize;		/* indir hash table size-1 */
1066	int	sd_on_journal;			/* Items on the journal list */
1067	int	sd_on_worklist;			/* Items on the worklist */
1068	int	sd_deps;			/* Total dependency count */
1069	int	sd_accdeps;			/* accumulated dep count */
1070	int	sd_req;				/* Wakeup when deps hits 0. */
1071	int	sd_flags;			/* comm with flushing thread */
1072	int	sd_cleanups;			/* Calls to cleanup */
1073	struct	thread *sd_flushtd;		/* thread handling flushing */
1074	TAILQ_ENTRY(mount_softdeps) sd_next;	/* List of softdep filesystem */
1075	struct	ufsmount *sd_ump;		/* our ufsmount structure */
1076	uint64_t sd_curdeps[D_LAST + 1];	/* count of current deps */
1077	struct	workhead sd_alldeps[D_LAST + 1];/* Lists of all deps */
1078};
1079/*
1080 * Flags for communicating with the syncer thread.
1081 */
1082#define FLUSH_EXIT	0x0001	/* time to exit */
1083#define FLUSH_CLEANUP	0x0002	/* need to clear out softdep structures */
1084#define	FLUSH_STARTING	0x0004	/* flush thread not yet started */
1085#define	FLUSH_RC_ACTIVE	0x0008	/* a thread is flushing the mount point */
1086#define	FLUSH_DI_ACTIVE	0x0010	/* a thread is processing delayed
1087				   inactivations */
1088
1089/*
1090 * Keep the old names from when these were in the ufsmount structure.
1091 */
1092#define	softdep_workitem_pending	um_softdep->sd_workitem_pending
1093#define	softdep_worklist_tail		um_softdep->sd_worklist_tail
1094#define	softdep_journal_pending		um_softdep->sd_journal_pending
1095#define	softdep_journal_tail		um_softdep->sd_journal_tail
1096#define	softdep_jblocks			um_softdep->sd_jblocks
1097#define	softdep_unlinked		um_softdep->sd_unlinked
1098#define	softdep_dirtycg			um_softdep->sd_dirtycg
1099#define	softdep_mkdirlisthd		um_softdep->sd_mkdirlisthd
1100#define	pagedep_hashtbl			um_softdep->sd_pdhash
1101#define	pagedep_hash_size		um_softdep->sd_pdhashsize
1102#define	pagedep_nextclean		um_softdep->sd_pdnextclean
1103#define	inodedep_hashtbl		um_softdep->sd_idhash
1104#define	inodedep_hash_size		um_softdep->sd_idhashsize
1105#define	inodedep_nextclean		um_softdep->sd_idnextclean
1106#define	newblk_hashtbl			um_softdep->sd_newblkhash
1107#define	newblk_hash_size		um_softdep->sd_newblkhashsize
1108#define	bmsafemap_hashtbl		um_softdep->sd_bmhash
1109#define	bmsafemap_hash_size		um_softdep->sd_bmhashsize
1110#define	indir_hashtbl			um_softdep->sd_indirhash
1111#define	indir_hash_size			um_softdep->sd_indirhashsize
1112#define	softdep_on_journal		um_softdep->sd_on_journal
1113#define	softdep_on_worklist		um_softdep->sd_on_worklist
1114#define	softdep_deps			um_softdep->sd_deps
1115#define	softdep_accdeps			um_softdep->sd_accdeps
1116#define	softdep_req			um_softdep->sd_req
1117#define	softdep_flags			um_softdep->sd_flags
1118#define	softdep_flushtd			um_softdep->sd_flushtd
1119#define	softdep_curdeps			um_softdep->sd_curdeps
1120#define	softdep_alldeps			um_softdep->sd_alldeps