1.. SPDX-License-Identifier: GPL-2.0 2.. _xfs_online_fsck_design: 3 4.. 5 Mapping of heading styles within this document: 6 Heading 1 uses "====" above and below 7 Heading 2 uses "====" 8 Heading 3 uses "----" 9 Heading 4 uses "````" 10 Heading 5 uses "^^^^" 11 Heading 6 uses "~~~~" 12 Heading 7 uses "...." 13 14 Sections are manually numbered because apparently that's what everyone 15 does in the kernel. 16 17====================== 18XFS Online Fsck Design 19====================== 20 21This document captures the design of the online filesystem check feature for 22XFS. 23The purpose of this document is threefold: 24 25- To help kernel distributors understand exactly what the XFS online fsck 26 feature is, and issues about which they should be aware. 27 28- To help people reading the code to familiarize themselves with the relevant 29 concepts and design points before they start digging into the code. 30 31- To help developers maintaining the system by capturing the reasons 32 supporting higher level decision making. 33 34As the online fsck code is merged, the links in this document to topic branches 35will be replaced with links to code. 36 37This document is licensed under the terms of the GNU Public License, v2. 38The primary author is Darrick J. Wong. 39 40This design document is split into seven parts. 41Part 1 defines what fsck tools are and the motivations for writing a new one. 42Parts 2 and 3 present a high level overview of how online fsck process works 43and how it is tested to ensure correct functionality. 44Part 4 discusses the user interface and the intended usage modes of the new 45program. 46Parts 5 and 6 show off the high level components and how they fit together, and 47then present case studies of how each repair function actually works. 48Part 7 sums up what has been discussed so far and speculates about what else 49might be built atop online fsck. 50 51.. contents:: Table of Contents 52 :local: 53 541. What is a Filesystem Check? 55============================== 56 57A Unix filesystem has four main responsibilities: 58 59- Provide a hierarchy of names through which application programs can associate 60 arbitrary blobs of data for any length of time, 61 62- Virtualize physical storage media across those names, and 63 64- Retrieve the named data blobs at any time. 65 66- Examine resource usage. 67 68Metadata directly supporting these functions (e.g. files, directories, space 69mappings) are sometimes called primary metadata. 70Secondary metadata (e.g. reverse mapping and directory parent pointers) support 71operations internal to the filesystem, such as internal consistency checking 72and reorganization. 73Summary metadata, as the name implies, condense information contained in 74primary metadata for performance reasons. 75 76The filesystem check (fsck) tool examines all the metadata in a filesystem 77to look for errors. 78In addition to looking for obvious metadata corruptions, fsck also 79cross-references different types of metadata records with each other to look 80for inconsistencies. 81People do not like losing data, so most fsck tools also contains some ability 82to correct any problems found. 83As a word of caution -- the primary goal of most Linux fsck tools is to restore 84the filesystem metadata to a consistent state, not to maximize the data 85recovered. 86That precedent will not be challenged here. 87 88Filesystems of the 20th century generally lacked any redundancy in the ondisk 89format, which means that fsck can only respond to errors by erasing files until 90errors are no longer detected. 91More recent filesystem designs contain enough redundancy in their metadata that 92it is now possible to regenerate data structures when non-catastrophic errors 93occur; this capability aids both strategies. 94 95+--------------------------------------------------------------------------+ 96| **Note**: | 97+--------------------------------------------------------------------------+ 98| System administrators avoid data loss by increasing the number of | 99| separate storage systems through the creation of backups; and they avoid | 100| downtime by increasing the redundancy of each storage system through the | 101| creation of RAID arrays. | 102| fsck tools address only the first problem. | 103+--------------------------------------------------------------------------+ 104 105TLDR; Show Me the Code! 106----------------------- 107 108Code is posted to the kernel.org git trees as follows: 109`kernel changes <https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-symlink>`_, 110`userspace changes <https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=scrub-media-scan-service>`_, and 111`QA test changes <https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfstests-dev.git/log/?h=repair-dirs>`_. 112Each kernel patchset adding an online repair function will use the same branch 113name across the kernel, xfsprogs, and fstests git repos. 114 115Existing Tools 116-------------- 117 118The online fsck tool described here will be the third tool in the history of 119XFS (on Linux) to check and repair filesystems. 120Two programs precede it: 121 122The first program, ``xfs_check``, was created as part of the XFS debugger 123(``xfs_db``) and can only be used with unmounted filesystems. 124It walks all metadata in the filesystem looking for inconsistencies in the 125metadata, though it lacks any ability to repair what it finds. 126Due to its high memory requirements and inability to repair things, this 127program is now deprecated and will not be discussed further. 128 129The second program, ``xfs_repair``, was created to be faster and more robust 130than the first program. 131Like its predecessor, it can only be used with unmounted filesystems. 132It uses extent-based in-memory data structures to reduce memory consumption, 133and tries to schedule readahead IO appropriately to reduce I/O waiting time 134while it scans the metadata of the entire filesystem. 135The most important feature of this tool is its ability to respond to 136inconsistencies in file metadata and directory tree by erasing things as needed 137to eliminate problems. 138Space usage metadata are rebuilt from the observed file metadata. 139 140Problem Statement 141----------------- 142 143The current XFS tools leave several problems unsolved: 144 1451. **User programs** suddenly **lose access** to the filesystem when unexpected 146 shutdowns occur as a result of silent corruptions in the metadata. 147 These occur **unpredictably** and often without warning. 148 1492. **Users** experience a **total loss of service** during the recovery period 150 after an **unexpected shutdown** occurs. 151 1523. **Users** experience a **total loss of service** if the filesystem is taken 153 offline to **look for problems** proactively. 154 1554. **Data owners** cannot **check the integrity** of their stored data without 156 reading all of it. 157 This may expose them to substantial billing costs when a linear media scan 158 performed by the storage system administrator might suffice. 159 1605. **System administrators** cannot **schedule** a maintenance window to deal 161 with corruptions if they **lack the means** to assess filesystem health 162 while the filesystem is online. 163 1646. **Fleet monitoring tools** cannot **automate periodic checks** of filesystem 165 health when doing so requires **manual intervention** and downtime. 166 1677. **Users** can be tricked into **doing things they do not desire** when 168 malicious actors **exploit quirks of Unicode** to place misleading names 169 in directories. 170 171Given this definition of the problems to be solved and the actors who would 172benefit, the proposed solution is a third fsck tool that acts on a running 173filesystem. 174 175This new third program has three components: an in-kernel facility to check 176metadata, an in-kernel facility to repair metadata, and a userspace driver 177program to drive fsck activity on a live filesystem. 178``xfs_scrub`` is the name of the driver program. 179The rest of this document presents the goals and use cases of the new fsck 180tool, describes its major design points in connection to those goals, and 181discusses the similarities and differences with existing tools. 182 183+--------------------------------------------------------------------------+ 184| **Note**: | 185+--------------------------------------------------------------------------+ 186| Throughout this document, the existing offline fsck tool can also be | 187| referred to by its current name "``xfs_repair``". | 188| The userspace driver program for the new online fsck tool can be | 189| referred to as "``xfs_scrub``". | 190| The kernel portion of online fsck that validates metadata is called | 191| "online scrub", and portion of the kernel that fixes metadata is called | 192| "online repair". | 193+--------------------------------------------------------------------------+ 194 195The naming hierarchy is broken up into objects known as directories and files 196and the physical space is split into pieces known as allocation groups. 197Sharding enables better performance on highly parallel systems and helps to 198contain the damage when corruptions occur. 199The division of the filesystem into principal objects (allocation groups and 200inodes) means that there are ample opportunities to perform targeted checks and 201repairs on a subset of the filesystem. 202 203While this is going on, other parts continue processing IO requests. 204Even if a piece of filesystem metadata can only be regenerated by scanning the 205entire system, the scan can still be done in the background while other file 206operations continue. 207 208In summary, online fsck takes advantage of resource sharding and redundant 209metadata to enable targeted checking and repair operations while the system 210is running. 211This capability will be coupled to automatic system management so that 212autonomous self-healing of XFS maximizes service availability. 213 2142. Theory of Operation 215====================== 216 217Because it is necessary for online fsck to lock and scan live metadata objects, 218online fsck consists of three separate code components. 219The first is the userspace driver program ``xfs_scrub``, which is responsible 220for identifying individual metadata items, scheduling work items for them, 221reacting to the outcomes appropriately, and reporting results to the system 222administrator. 223The second and third are in the kernel, which implements functions to check 224and repair each type of online fsck work item. 225 226+------------------------------------------------------------------+ 227| **Note**: | 228+------------------------------------------------------------------+ 229| For brevity, this document shortens the phrase "online fsck work | 230| item" to "scrub item". | 231+------------------------------------------------------------------+ 232 233Scrub item types are delineated in a manner consistent with the Unix design 234philosophy, which is to say that each item should handle one aspect of a 235metadata structure, and handle it well. 236 237Scope 238----- 239 240In principle, online fsck should be able to check and to repair everything that 241the offline fsck program can handle. 242However, online fsck cannot be running 100% of the time, which means that 243latent errors may creep in after a scrub completes. 244If these errors cause the next mount to fail, offline fsck is the only 245solution. 246This limitation means that maintenance of the offline fsck tool will continue. 247A second limitation of online fsck is that it must follow the same resource 248sharing and lock acquisition rules as the regular filesystem. 249This means that scrub cannot take *any* shortcuts to save time, because doing 250so could lead to concurrency problems. 251In other words, online fsck is not a complete replacement for offline fsck, and 252a complete run of online fsck may take longer than online fsck. 253However, both of these limitations are acceptable tradeoffs to satisfy the 254different motivations of online fsck, which are to **minimize system downtime** 255and to **increase predictability of operation**. 256 257.. _scrubphases: 258 259Phases of Work 260-------------- 261 262The userspace driver program ``xfs_scrub`` splits the work of checking and 263repairing an entire filesystem into seven phases. 264Each phase concentrates on checking specific types of scrub items and depends 265on the success of all previous phases. 266The seven phases are as follows: 267 2681. Collect geometry information about the mounted filesystem and computer, 269 discover the online fsck capabilities of the kernel, and open the 270 underlying storage devices. 271 2722. Check allocation group metadata, all realtime volume metadata, and all quota 273 files. 274 Each metadata structure is scheduled as a separate scrub item. 275 If corruption is found in the inode header or inode btree and ``xfs_scrub`` 276 is permitted to perform repairs, then those scrub items are repaired to 277 prepare for phase 3. 278 Repairs are implemented by using the information in the scrub item to 279 resubmit the kernel scrub call with the repair flag enabled; this is 280 discussed in the next section. 281 Optimizations and all other repairs are deferred to phase 4. 282 2833. Check all metadata of every file in the filesystem. 284 Each metadata structure is also scheduled as a separate scrub item. 285 If repairs are needed and ``xfs_scrub`` is permitted to perform repairs, 286 and there were no problems detected during phase 2, then those scrub items 287 are repaired immediately. 288 Optimizations, deferred repairs, and unsuccessful repairs are deferred to 289 phase 4. 290 2914. All remaining repairs and scheduled optimizations are performed during this 292 phase, if the caller permits them. 293 Before starting repairs, the summary counters are checked and any necessary 294 repairs are performed so that subsequent repairs will not fail the resource 295 reservation step due to wildly incorrect summary counters. 296 Unsuccessful repairs are requeued as long as forward progress on repairs is 297 made somewhere in the filesystem. 298 Free space in the filesystem is trimmed at the end of phase 4 if the 299 filesystem is clean. 300 3015. By the start of this phase, all primary and secondary filesystem metadata 302 must be correct. 303 Summary counters such as the free space counts and quota resource counts 304 are checked and corrected. 305 Directory entry names and extended attribute names are checked for 306 suspicious entries such as control characters or confusing Unicode sequences 307 appearing in names. 308 3096. If the caller asks for a media scan, read all allocated and written data 310 file extents in the filesystem. 311 The ability to use hardware-assisted data file integrity checking is new 312 to online fsck; neither of the previous tools have this capability. 313 If media errors occur, they will be mapped to the owning files and reported. 314 3157. Re-check the summary counters and presents the caller with a summary of 316 space usage and file counts. 317 318This allocation of responsibilities will be :ref:`revisited <scrubcheck>` 319later in this document. 320 321Steps for Each Scrub Item 322------------------------- 323 324The kernel scrub code uses a three-step strategy for checking and repairing 325the one aspect of a metadata object represented by a scrub item: 326 3271. The scrub item of interest is checked for corruptions; opportunities for 328 optimization; and for values that are directly controlled by the system 329 administrator but look suspicious. 330 If the item is not corrupt or does not need optimization, resource are 331 released and the positive scan results are returned to userspace. 332 If the item is corrupt or could be optimized but the caller does not permit 333 this, resources are released and the negative scan results are returned to 334 userspace. 335 Otherwise, the kernel moves on to the second step. 336 3372. The repair function is called to rebuild the data structure. 338 Repair functions generally choose rebuild a structure from other metadata 339 rather than try to salvage the existing structure. 340 If the repair fails, the scan results from the first step are returned to 341 userspace. 342 Otherwise, the kernel moves on to the third step. 343 3443. In the third step, the kernel runs the same checks over the new metadata 345 item to assess the efficacy of the repairs. 346 The results of the reassessment are returned to userspace. 347 348Classification of Metadata 349-------------------------- 350 351Each type of metadata object (and therefore each type of scrub item) is 352classified as follows: 353 354Primary Metadata 355```````````````` 356 357Metadata structures in this category should be most familiar to filesystem 358users either because they are directly created by the user or they index 359objects created by the user 360Most filesystem objects fall into this class: 361 362- Free space and reference count information 363 364- Inode records and indexes 365 366- Storage mapping information for file data 367 368- Directories 369 370- Extended attributes 371 372- Symbolic links 373 374- Quota limits 375 376Scrub obeys the same rules as regular filesystem accesses for resource and lock 377acquisition. 378 379Primary metadata objects are the simplest for scrub to process. 380The principal filesystem object (either an allocation group or an inode) that 381owns the item being scrubbed is locked to guard against concurrent updates. 382The check function examines every record associated with the type for obvious 383errors and cross-references healthy records against other metadata to look for 384inconsistencies. 385Repairs for this class of scrub item are simple, since the repair function 386starts by holding all the resources acquired in the previous step. 387The repair function scans available metadata as needed to record all the 388observations needed to complete the structure. 389Next, it stages the observations in a new ondisk structure and commits it 390atomically to complete the repair. 391Finally, the storage from the old data structure are carefully reaped. 392 393Because ``xfs_scrub`` locks a primary object for the duration of the repair, 394this is effectively an offline repair operation performed on a subset of the 395filesystem. 396This minimizes the complexity of the repair code because it is not necessary to 397handle concurrent updates from other threads, nor is it necessary to access 398any other part of the filesystem. 399As a result, indexed structures can be rebuilt very quickly, and programs 400trying to access the damaged structure will be blocked until repairs complete. 401The only infrastructure needed by the repair code are the staging area for 402observations and a means to write new structures to disk. 403Despite these limitations, the advantage that online repair holds is clear: 404targeted work on individual shards of the filesystem avoids total loss of 405service. 406 407This mechanism is described in section 2.1 ("Off-Line Algorithm") of 408V. Srinivasan and M. J. Carey, `"Performance of On-Line Index Construction 409Algorithms" <https://minds.wisconsin.edu/bitstream/handle/1793/59524/TR1047.pdf>`_, 410*Extending Database Technology*, pp. 293-309, 1992. 411 412Most primary metadata repair functions stage their intermediate results in an 413in-memory array prior to formatting the new ondisk structure, which is very 414similar to the list-based algorithm discussed in section 2.3 ("List-Based 415Algorithms") of Srinivasan. 416However, any data structure builder that maintains a resource lock for the 417duration of the repair is *always* an offline algorithm. 418 419.. _secondary_metadata: 420 421Secondary Metadata 422`````````````````` 423 424Metadata structures in this category reflect records found in primary metadata, 425but are only needed for online fsck or for reorganization of the filesystem. 426 427Secondary metadata include: 428 429- Reverse mapping information 430 431- Directory parent pointers 432 433This class of metadata is difficult for scrub to process because scrub attaches 434to the secondary object but needs to check primary metadata, which runs counter 435to the usual order of resource acquisition. 436Frequently, this means that full filesystems scans are necessary to rebuild the 437metadata. 438Check functions can be limited in scope to reduce runtime. 439Repairs, however, require a full scan of primary metadata, which can take a 440long time to complete. 441Under these conditions, ``xfs_scrub`` cannot lock resources for the entire 442duration of the repair. 443 444Instead, repair functions set up an in-memory staging structure to store 445observations. 446Depending on the requirements of the specific repair function, the staging 447index will either have the same format as the ondisk structure or a design 448specific to that repair function. 449The next step is to release all locks and start the filesystem scan. 450When the repair scanner needs to record an observation, the staging data are 451locked long enough to apply the update. 452While the filesystem scan is in progress, the repair function hooks the 453filesystem so that it can apply pending filesystem updates to the staging 454information. 455Once the scan is done, the owning object is re-locked, the live data is used to 456write a new ondisk structure, and the repairs are committed atomically. 457The hooks are disabled and the staging staging area is freed. 458Finally, the storage from the old data structure are carefully reaped. 459 460Introducing concurrency helps online repair avoid various locking problems, but 461comes at a high cost to code complexity. 462Live filesystem code has to be hooked so that the repair function can observe 463updates in progress. 464The staging area has to become a fully functional parallel structure so that 465updates can be merged from the hooks. 466Finally, the hook, the filesystem scan, and the inode locking model must be 467sufficiently well integrated that a hook event can decide if a given update 468should be applied to the staging structure. 469 470In theory, the scrub implementation could apply these same techniques for 471primary metadata, but doing so would make it massively more complex and less 472performant. 473Programs attempting to access the damaged structures are not blocked from 474operation, which may cause application failure or an unplanned filesystem 475shutdown. 476 477Inspiration for the secondary metadata repair strategy was drawn from section 4782.4 of Srinivasan above, and sections 2 ("NSF: Inded Build Without Side-File") 479and 3.1.1 ("Duplicate Key Insert Problem") in C. Mohan, `"Algorithms for 480Creating Indexes for Very Large Tables Without Quiescing Updates" 481<https://dl.acm.org/doi/10.1145/130283.130337>`_, 1992. 482 483The sidecar index mentioned above bears some resemblance to the side file 484method mentioned in Srinivasan and Mohan. 485Their method consists of an index builder that extracts relevant record data to 486build the new structure as quickly as possible; and an auxiliary structure that 487captures all updates that would be committed to the index by other threads were 488the new index already online. 489After the index building scan finishes, the updates recorded in the side file 490are applied to the new index. 491To avoid conflicts between the index builder and other writer threads, the 492builder maintains a publicly visible cursor that tracks the progress of the 493scan through the record space. 494To avoid duplication of work between the side file and the index builder, side 495file updates are elided when the record ID for the update is greater than the 496cursor position within the record ID space. 497 498To minimize changes to the rest of the codebase, XFS online repair keeps the 499replacement index hidden until it's completely ready to go. 500In other words, there is no attempt to expose the keyspace of the new index 501while repair is running. 502The complexity of such an approach would be very high and perhaps more 503appropriate to building *new* indices. 504 505**Future Work Question**: Can the full scan and live update code used to 506facilitate a repair also be used to implement a comprehensive check? 507 508*Answer*: In theory, yes. Check would be much stronger if each scrub function 509employed these live scans to build a shadow copy of the metadata and then 510compared the shadow records to the ondisk records. 511However, doing that is a fair amount more work than what the checking functions 512do now. 513The live scans and hooks were developed much later. 514That in turn increases the runtime of those scrub functions. 515 516Summary Information 517``````````````````` 518 519Metadata structures in this last category summarize the contents of primary 520metadata records. 521These are often used to speed up resource usage queries, and are many times 522smaller than the primary metadata which they represent. 523 524Examples of summary information include: 525 526- Summary counts of free space and inodes 527 528- File link counts from directories 529 530- Quota resource usage counts 531 532Check and repair require full filesystem scans, but resource and lock 533acquisition follow the same paths as regular filesystem accesses. 534 535The superblock summary counters have special requirements due to the underlying 536implementation of the incore counters, and will be treated separately. 537Check and repair of the other types of summary counters (quota resource counts 538and file link counts) employ the same filesystem scanning and hooking 539techniques as outlined above, but because the underlying data are sets of 540integer counters, the staging data need not be a fully functional mirror of the 541ondisk structure. 542 543Inspiration for quota and file link count repair strategies were drawn from 544sections 2.12 ("Online Index Operations") through 2.14 ("Incremental View 545Maintenance") of G. Graefe, `"Concurrent Queries and Updates in Summary Views 546and Their Indexes" 547<http://www.odbms.org/wp-content/uploads/2014/06/Increment-locks.pdf>`_, 2011. 548 549Since quotas are non-negative integer counts of resource usage, online 550quotacheck can use the incremental view deltas described in section 2.14 to 551track pending changes to the block and inode usage counts in each transaction, 552and commit those changes to a dquot side file when the transaction commits. 553Delta tracking is necessary for dquots because the index builder scans inodes, 554whereas the data structure being rebuilt is an index of dquots. 555Link count checking combines the view deltas and commit step into one because 556it sets attributes of the objects being scanned instead of writing them to a 557separate data structure. 558Each online fsck function will be discussed as case studies later in this 559document. 560 561Risk Management 562--------------- 563 564During the development of online fsck, several risk factors were identified 565that may make the feature unsuitable for certain distributors and users. 566Steps can be taken to mitigate or eliminate those risks, though at a cost to 567functionality. 568 569- **Decreased performance**: Adding metadata indices to the filesystem 570 increases the time cost of persisting changes to disk, and the reverse space 571 mapping and directory parent pointers are no exception. 572 System administrators who require the maximum performance can disable the 573 reverse mapping features at format time, though this choice dramatically 574 reduces the ability of online fsck to find inconsistencies and repair them. 575 576- **Incorrect repairs**: As with all software, there might be defects in the 577 software that result in incorrect repairs being written to the filesystem. 578 Systematic fuzz testing (detailed in the next section) is employed by the 579 authors to find bugs early, but it might not catch everything. 580 The kernel build system provides Kconfig options (``CONFIG_XFS_ONLINE_SCRUB`` 581 and ``CONFIG_XFS_ONLINE_REPAIR``) to enable distributors to choose not to 582 accept this risk. 583 The xfsprogs build system has a configure option (``--enable-scrub=no``) that 584 disables building of the ``xfs_scrub`` binary, though this is not a risk 585 mitigation if the kernel functionality remains enabled. 586 587- **Inability to repair**: Sometimes, a filesystem is too badly damaged to be 588 repairable. 589 If the keyspaces of several metadata indices overlap in some manner but a 590 coherent narrative cannot be formed from records collected, then the repair 591 fails. 592 To reduce the chance that a repair will fail with a dirty transaction and 593 render the filesystem unusable, the online repair functions have been 594 designed to stage and validate all new records before committing the new 595 structure. 596 597- **Misbehavior**: Online fsck requires many privileges -- raw IO to block 598 devices, opening files by handle, ignoring Unix discretionary access control, 599 and the ability to perform administrative changes. 600 Running this automatically in the background scares people, so the systemd 601 background service is configured to run with only the privileges required. 602 Obviously, this cannot address certain problems like the kernel crashing or 603 deadlocking, but it should be sufficient to prevent the scrub process from 604 escaping and reconfiguring the system. 605 The cron job does not have this protection. 606 607- **Fuzz Kiddiez**: There are many people now who seem to think that running 608 automated fuzz testing of ondisk artifacts to find mischievous behavior and 609 spraying exploit code onto the public mailing list for instant zero-day 610 disclosure is somehow of some social benefit. 611 In the view of this author, the benefit is realized only when the fuzz 612 operators help to **fix** the flaws, but this opinion apparently is not 613 widely shared among security "researchers". 614 The XFS maintainers' continuing ability to manage these events presents an 615 ongoing risk to the stability of the development process. 616 Automated testing should front-load some of the risk while the feature is 617 considered EXPERIMENTAL. 618 619Many of these risks are inherent to software programming. 620Despite this, it is hoped that this new functionality will prove useful in 621reducing unexpected downtime. 622 6233. Testing Plan 624=============== 625 626As stated before, fsck tools have three main goals: 627 6281. Detect inconsistencies in the metadata; 629 6302. Eliminate those inconsistencies; and 631 6323. Minimize further loss of data. 633 634Demonstrations of correct operation are necessary to build users' confidence 635that the software behaves within expectations. 636Unfortunately, it was not really feasible to perform regular exhaustive testing 637of every aspect of a fsck tool until the introduction of low-cost virtual 638machines with high-IOPS storage. 639With ample hardware availability in mind, the testing strategy for the online 640fsck project involves differential analysis against the existing fsck tools and 641systematic testing of every attribute of every type of metadata object. 642Testing can be split into four major categories, as discussed below. 643 644Integrated Testing with fstests 645------------------------------- 646 647The primary goal of any free software QA effort is to make testing as 648inexpensive and widespread as possible to maximize the scaling advantages of 649community. 650In other words, testing should maximize the breadth of filesystem configuration 651scenarios and hardware setups. 652This improves code quality by enabling the authors of online fsck to find and 653fix bugs early, and helps developers of new features to find integration 654issues earlier in their development effort. 655 656The Linux filesystem community shares a common QA testing suite, 657`fstests <https://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git/>`_, for 658functional and regression testing. 659Even before development work began on online fsck, fstests (when run on XFS) 660would run both the ``xfs_check`` and ``xfs_repair -n`` commands on the test and 661scratch filesystems between each test. 662This provides a level of assurance that the kernel and the fsck tools stay in 663alignment about what constitutes consistent metadata. 664During development of the online checking code, fstests was modified to run 665``xfs_scrub -n`` between each test to ensure that the new checking code 666produces the same results as the two existing fsck tools. 667 668To start development of online repair, fstests was modified to run 669``xfs_repair`` to rebuild the filesystem's metadata indices between tests. 670This ensures that offline repair does not crash, leave a corrupt filesystem 671after it exists, or trigger complaints from the online check. 672This also established a baseline for what can and cannot be repaired offline. 673To complete the first phase of development of online repair, fstests was 674modified to be able to run ``xfs_scrub`` in a "force rebuild" mode. 675This enables a comparison of the effectiveness of online repair as compared to 676the existing offline repair tools. 677 678General Fuzz Testing of Metadata Blocks 679--------------------------------------- 680 681XFS benefits greatly from having a very robust debugging tool, ``xfs_db``. 682 683Before development of online fsck even began, a set of fstests were created 684to test the rather common fault that entire metadata blocks get corrupted. 685This required the creation of fstests library code that can create a filesystem 686containing every possible type of metadata object. 687Next, individual test cases were created to create a test filesystem, identify 688a single block of a specific type of metadata object, trash it with the 689existing ``blocktrash`` command in ``xfs_db``, and test the reaction of a 690particular metadata validation strategy. 691 692This earlier test suite enabled XFS developers to test the ability of the 693in-kernel validation functions and the ability of the offline fsck tool to 694detect and eliminate the inconsistent metadata. 695This part of the test suite was extended to cover online fsck in exactly the 696same manner. 697 698In other words, for a given fstests filesystem configuration: 699 700* For each metadata object existing on the filesystem: 701 702 * Write garbage to it 703 704 * Test the reactions of: 705 706 1. The kernel verifiers to stop obviously bad metadata 707 2. Offline repair (``xfs_repair``) to detect and fix 708 3. Online repair (``xfs_scrub``) to detect and fix 709 710Targeted Fuzz Testing of Metadata Records 711----------------------------------------- 712 713The testing plan for online fsck includes extending the existing fs testing 714infrastructure to provide a much more powerful facility: targeted fuzz testing 715of every metadata field of every metadata object in the filesystem. 716``xfs_db`` can modify every field of every metadata structure in every 717block in the filesystem to simulate the effects of memory corruption and 718software bugs. 719Given that fstests already contains the ability to create a filesystem 720containing every metadata format known to the filesystem, ``xfs_db`` can be 721used to perform exhaustive fuzz testing! 722 723For a given fstests filesystem configuration: 724 725* For each metadata object existing on the filesystem... 726 727 * For each record inside that metadata object... 728 729 * For each field inside that record... 730 731 * For each conceivable type of transformation that can be applied to a bit field... 732 733 1. Clear all bits 734 2. Set all bits 735 3. Toggle the most significant bit 736 4. Toggle the middle bit 737 5. Toggle the least significant bit 738 6. Add a small quantity 739 7. Subtract a small quantity 740 8. Randomize the contents 741 742 * ...test the reactions of: 743 744 1. The kernel verifiers to stop obviously bad metadata 745 2. Offline checking (``xfs_repair -n``) 746 3. Offline repair (``xfs_repair``) 747 4. Online checking (``xfs_scrub -n``) 748 5. Online repair (``xfs_scrub``) 749 6. Both repair tools (``xfs_scrub`` and then ``xfs_repair`` if online repair doesn't succeed) 750 751This is quite the combinatoric explosion! 752 753Fortunately, having this much test coverage makes it easy for XFS developers to 754check the responses of XFS' fsck tools. 755Since the introduction of the fuzz testing framework, these tests have been 756used to discover incorrect repair code and missing functionality for entire 757classes of metadata objects in ``xfs_repair``. 758The enhanced testing was used to finalize the deprecation of ``xfs_check`` by 759confirming that ``xfs_repair`` could detect at least as many corruptions as 760the older tool. 761 762These tests have been very valuable for ``xfs_scrub`` in the same ways -- they 763allow the online fsck developers to compare online fsck against offline fsck, 764and they enable XFS developers to find deficiencies in the code base. 765 766Proposed patchsets include 767`general fuzzer improvements 768<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfstests-dev.git/log/?h=fuzzer-improvements>`_, 769`fuzzing baselines 770<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfstests-dev.git/log/?h=fuzz-baseline>`_, 771and `improvements in fuzz testing comprehensiveness 772<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfstests-dev.git/log/?h=more-fuzz-testing>`_. 773 774Stress Testing 775-------------- 776 777A unique requirement to online fsck is the ability to operate on a filesystem 778concurrently with regular workloads. 779Although it is of course impossible to run ``xfs_scrub`` with *zero* observable 780impact on the running system, the online repair code should never introduce 781inconsistencies into the filesystem metadata, and regular workloads should 782never notice resource starvation. 783To verify that these conditions are being met, fstests has been enhanced in 784the following ways: 785 786* For each scrub item type, create a test to exercise checking that item type 787 while running ``fsstress``. 788* For each scrub item type, create a test to exercise repairing that item type 789 while running ``fsstress``. 790* Race ``fsstress`` and ``xfs_scrub -n`` to ensure that checking the whole 791 filesystem doesn't cause problems. 792* Race ``fsstress`` and ``xfs_scrub`` in force-rebuild mode to ensure that 793 force-repairing the whole filesystem doesn't cause problems. 794* Race ``xfs_scrub`` in check and force-repair mode against ``fsstress`` while 795 freezing and thawing the filesystem. 796* Race ``xfs_scrub`` in check and force-repair mode against ``fsstress`` while 797 remounting the filesystem read-only and read-write. 798* The same, but running ``fsx`` instead of ``fsstress``. (Not done yet?) 799 800Success is defined by the ability to run all of these tests without observing 801any unexpected filesystem shutdowns due to corrupted metadata, kernel hang 802check warnings, or any other sort of mischief. 803 804Proposed patchsets include `general stress testing 805<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfstests-dev.git/log/?h=race-scrub-and-mount-state-changes>`_ 806and the `evolution of existing per-function stress testing 807<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfstests-dev.git/log/?h=refactor-scrub-stress>`_. 808 8094. User Interface 810================= 811 812The primary user of online fsck is the system administrator, just like offline 813repair. 814Online fsck presents two modes of operation to administrators: 815A foreground CLI process for online fsck on demand, and a background service 816that performs autonomous checking and repair. 817 818Checking on Demand 819------------------ 820 821For administrators who want the absolute freshest information about the 822metadata in a filesystem, ``xfs_scrub`` can be run as a foreground process on 823a command line. 824The program checks every piece of metadata in the filesystem while the 825administrator waits for the results to be reported, just like the existing 826``xfs_repair`` tool. 827Both tools share a ``-n`` option to perform a read-only scan, and a ``-v`` 828option to increase the verbosity of the information reported. 829 830A new feature of ``xfs_scrub`` is the ``-x`` option, which employs the error 831correction capabilities of the hardware to check data file contents. 832The media scan is not enabled by default because it may dramatically increase 833program runtime and consume a lot of bandwidth on older storage hardware. 834 835The output of a foreground invocation is captured in the system log. 836 837The ``xfs_scrub_all`` program walks the list of mounted filesystems and 838initiates ``xfs_scrub`` for each of them in parallel. 839It serializes scans for any filesystems that resolve to the same top level 840kernel block device to prevent resource overconsumption. 841 842Background Service 843------------------ 844 845To reduce the workload of system administrators, the ``xfs_scrub`` package 846provides a suite of `systemd <https://systemd.io/>`_ timers and services that 847run online fsck automatically on weekends by default. 848The background service configures scrub to run with as little privilege as 849possible, the lowest CPU and IO priority, and in a CPU-constrained single 850threaded mode. 851This can be tuned by the systemd administrator at any time to suit the latency 852and throughput requirements of customer workloads. 853 854The output of the background service is also captured in the system log. 855If desired, reports of failures (either due to inconsistencies or mere runtime 856errors) can be emailed automatically by setting the ``EMAIL_ADDR`` environment 857variable in the following service files: 858 859* ``xfs_scrub_fail@.service`` 860* ``xfs_scrub_media_fail@.service`` 861* ``xfs_scrub_all_fail.service`` 862 863The decision to enable the background scan is left to the system administrator. 864This can be done by enabling either of the following services: 865 866* ``xfs_scrub_all.timer`` on systemd systems 867* ``xfs_scrub_all.cron`` on non-systemd systems 868 869This automatic weekly scan is configured out of the box to perform an 870additional media scan of all file data once per month. 871This is less foolproof than, say, storing file data block checksums, but much 872more performant if application software provides its own integrity checking, 873redundancy can be provided elsewhere above the filesystem, or the storage 874device's integrity guarantees are deemed sufficient. 875 876The systemd unit file definitions have been subjected to a security audit 877(as of systemd 249) to ensure that the xfs_scrub processes have as little 878access to the rest of the system as possible. 879This was performed via ``systemd-analyze security``, after which privileges 880were restricted to the minimum required, sandboxing was set up to the maximal 881extent possible with sandboxing and system call filtering; and access to the 882filesystem tree was restricted to the minimum needed to start the program and 883access the filesystem being scanned. 884The service definition files restrict CPU usage to 80% of one CPU core, and 885apply as nice of a priority to IO and CPU scheduling as possible. 886This measure was taken to minimize delays in the rest of the filesystem. 887No such hardening has been performed for the cron job. 888 889Proposed patchset: 890`Enabling the xfs_scrub background service 891<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=scrub-media-scan-service>`_. 892 893Health Reporting 894---------------- 895 896XFS caches a summary of each filesystem's health status in memory. 897The information is updated whenever ``xfs_scrub`` is run, or whenever 898inconsistencies are detected in the filesystem metadata during regular 899operations. 900System administrators should use the ``health`` command of ``xfs_spaceman`` to 901download this information into a human-readable format. 902If problems have been observed, the administrator can schedule a reduced 903service window to run the online repair tool to correct the problem. 904Failing that, the administrator can decide to schedule a maintenance window to 905run the traditional offline repair tool to correct the problem. 906 907**Future Work Question**: Should the health reporting integrate with the new 908inotify fs error notification system? 909Would it be helpful for sysadmins to have a daemon to listen for corruption 910notifications and initiate a repair? 911 912*Answer*: These questions remain unanswered, but should be a part of the 913conversation with early adopters and potential downstream users of XFS. 914 915Proposed patchsets include 916`wiring up health reports to correction returns 917<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=corruption-health-reports>`_ 918and 919`preservation of sickness info during memory reclaim 920<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=indirect-health-reporting>`_. 921 9225. Kernel Algorithms and Data Structures 923======================================== 924 925This section discusses the key algorithms and data structures of the kernel 926code that provide the ability to check and repair metadata while the system 927is running. 928The first chapters in this section reveal the pieces that provide the 929foundation for checking metadata. 930The remainder of this section presents the mechanisms through which XFS 931regenerates itself. 932 933Self Describing Metadata 934------------------------ 935 936Starting with XFS version 5 in 2012, XFS updated the format of nearly every 937ondisk block header to record a magic number, a checksum, a universally 938"unique" identifier (UUID), an owner code, the ondisk address of the block, 939and a log sequence number. 940When loading a block buffer from disk, the magic number, UUID, owner, and 941ondisk address confirm that the retrieved block matches the specific owner of 942the current filesystem, and that the information contained in the block is 943supposed to be found at the ondisk address. 944The first three components enable checking tools to disregard alleged metadata 945that doesn't belong to the filesystem, and the fourth component enables the 946filesystem to detect lost writes. 947 948Whenever a file system operation modifies a block, the change is submitted 949to the log as part of a transaction. 950The log then processes these transactions marking them done once they are 951safely persisted to storage. 952The logging code maintains the checksum and the log sequence number of the last 953transactional update. 954Checksums are useful for detecting torn writes and other discrepancies that can 955be introduced between the computer and its storage devices. 956Sequence number tracking enables log recovery to avoid applying out of date 957log updates to the filesystem. 958 959These two features improve overall runtime resiliency by providing a means for 960the filesystem to detect obvious corruption when reading metadata blocks from 961disk, but these buffer verifiers cannot provide any consistency checking 962between metadata structures. 963 964For more information, please see the documentation for 965Documentation/filesystems/xfs/xfs-self-describing-metadata.rst 966 967Reverse Mapping 968--------------- 969 970The original design of XFS (circa 1993) is an improvement upon 1980s Unix 971filesystem design. 972In those days, storage density was expensive, CPU time was scarce, and 973excessive seek time could kill performance. 974For performance reasons, filesystem authors were reluctant to add redundancy to 975the filesystem, even at the cost of data integrity. 976Filesystems designers in the early 21st century choose different strategies to 977increase internal redundancy -- either storing nearly identical copies of 978metadata, or more space-efficient encoding techniques. 979 980For XFS, a different redundancy strategy was chosen to modernize the design: 981a secondary space usage index that maps allocated disk extents back to their 982owners. 983By adding a new index, the filesystem retains most of its ability to scale 984well to heavily threaded workloads involving large datasets, since the primary 985file metadata (the directory tree, the file block map, and the allocation 986groups) remain unchanged. 987Like any system that improves redundancy, the reverse-mapping feature increases 988overhead costs for space mapping activities. 989However, it has two critical advantages: first, the reverse index is key to 990enabling online fsck and other requested functionality such as free space 991defragmentation, better media failure reporting, and filesystem shrinking. 992Second, the different ondisk storage format of the reverse mapping btree 993defeats device-level deduplication because the filesystem requires real 994redundancy. 995 996+--------------------------------------------------------------------------+ 997| **Sidebar**: | 998+--------------------------------------------------------------------------+ 999| A criticism of adding the secondary index is that it does nothing to | 1000| improve the robustness of user data storage itself. | 1001| This is a valid point, but adding a new index for file data block | 1002| checksums increases write amplification by turning data overwrites into | 1003| copy-writes, which age the filesystem prematurely. | 1004| In keeping with thirty years of precedent, users who want file data | 1005| integrity can supply as powerful a solution as they require. | 1006| As for metadata, the complexity of adding a new secondary index of space | 1007| usage is much less than adding volume management and storage device | 1008| mirroring to XFS itself. | 1009| Perfection of RAID and volume management are best left to existing | 1010| layers in the kernel. | 1011+--------------------------------------------------------------------------+ 1012 1013The information captured in a reverse space mapping record is as follows: 1014 1015.. code-block:: c 1016 1017 struct xfs_rmap_irec { 1018 xfs_agblock_t rm_startblock; /* extent start block */ 1019 xfs_extlen_t rm_blockcount; /* extent length */ 1020 uint64_t rm_owner; /* extent owner */ 1021 uint64_t rm_offset; /* offset within the owner */ 1022 unsigned int rm_flags; /* state flags */ 1023 }; 1024 1025The first two fields capture the location and size of the physical space, 1026in units of filesystem blocks. 1027The owner field tells scrub which metadata structure or file inode have been 1028assigned this space. 1029For space allocated to files, the offset field tells scrub where the space was 1030mapped within the file fork. 1031Finally, the flags field provides extra information about the space usage -- 1032is this an attribute fork extent? A file mapping btree extent? Or an 1033unwritten data extent? 1034 1035Online filesystem checking judges the consistency of each primary metadata 1036record by comparing its information against all other space indices. 1037The reverse mapping index plays a key role in the consistency checking process 1038because it contains a centralized alternate copy of all space allocation 1039information. 1040Program runtime and ease of resource acquisition are the only real limits to 1041what online checking can consult. 1042For example, a file data extent mapping can be checked against: 1043 1044* The absence of an entry in the free space information. 1045* The absence of an entry in the inode index. 1046* The absence of an entry in the reference count data if the file is not 1047 marked as having shared extents. 1048* The correspondence of an entry in the reverse mapping information. 1049 1050There are several observations to make about reverse mapping indices: 1051 10521. Reverse mappings can provide a positive affirmation of correctness if any of 1053 the above primary metadata are in doubt. 1054 The checking code for most primary metadata follows a path similar to the 1055 one outlined above. 1056 10572. Proving the consistency of secondary metadata with the primary metadata is 1058 difficult because that requires a full scan of all primary space metadata, 1059 which is very time intensive. 1060 For example, checking a reverse mapping record for a file extent mapping 1061 btree block requires locking the file and searching the entire btree to 1062 confirm the block. 1063 Instead, scrub relies on rigorous cross-referencing during the primary space 1064 mapping structure checks. 1065 10663. Consistency scans must use non-blocking lock acquisition primitives if the 1067 required locking order is not the same order used by regular filesystem 1068 operations. 1069 For example, if the filesystem normally takes a file ILOCK before taking 1070 the AGF buffer lock but scrub wants to take a file ILOCK while holding 1071 an AGF buffer lock, scrub cannot block on that second acquisition. 1072 This means that forward progress during this part of a scan of the reverse 1073 mapping data cannot be guaranteed if system load is heavy. 1074 1075In summary, reverse mappings play a key role in reconstruction of primary 1076metadata. 1077The details of how these records are staged, written to disk, and committed 1078into the filesystem are covered in subsequent sections. 1079 1080Checking and Cross-Referencing 1081------------------------------ 1082 1083The first step of checking a metadata structure is to examine every record 1084contained within the structure and its relationship with the rest of the 1085system. 1086XFS contains multiple layers of checking to try to prevent inconsistent 1087metadata from wreaking havoc on the system. 1088Each of these layers contributes information that helps the kernel to make 1089three decisions about the health of a metadata structure: 1090 1091- Is a part of this structure obviously corrupt (``XFS_SCRUB_OFLAG_CORRUPT``) ? 1092- Is this structure inconsistent with the rest of the system 1093 (``XFS_SCRUB_OFLAG_XCORRUPT``) ? 1094- Is there so much damage around the filesystem that cross-referencing is not 1095 possible (``XFS_SCRUB_OFLAG_XFAIL``) ? 1096- Can the structure be optimized to improve performance or reduce the size of 1097 metadata (``XFS_SCRUB_OFLAG_PREEN``) ? 1098- Does the structure contain data that is not inconsistent but deserves review 1099 by the system administrator (``XFS_SCRUB_OFLAG_WARNING``) ? 1100 1101The following sections describe how the metadata scrubbing process works. 1102 1103Metadata Buffer Verification 1104```````````````````````````` 1105 1106The lowest layer of metadata protection in XFS are the metadata verifiers built 1107into the buffer cache. 1108These functions perform inexpensive internal consistency checking of the block 1109itself, and answer these questions: 1110 1111- Does the block belong to this filesystem? 1112 1113- Does the block belong to the structure that asked for the read? 1114 This assumes that metadata blocks only have one owner, which is always true 1115 in XFS. 1116 1117- Is the type of data stored in the block within a reasonable range of what 1118 scrub is expecting? 1119 1120- Does the physical location of the block match the location it was read from? 1121 1122- Does the block checksum match the data? 1123 1124The scope of the protections here are very limited -- verifiers can only 1125establish that the filesystem code is reasonably free of gross corruption bugs 1126and that the storage system is reasonably competent at retrieval. 1127Corruption problems observed at runtime cause the generation of health reports, 1128failed system calls, and in the extreme case, filesystem shutdowns if the 1129corrupt metadata force the cancellation of a dirty transaction. 1130 1131Every online fsck scrubbing function is expected to read every ondisk metadata 1132block of a structure in the course of checking the structure. 1133Corruption problems observed during a check are immediately reported to 1134userspace as corruption; during a cross-reference, they are reported as a 1135failure to cross-reference once the full examination is complete. 1136Reads satisfied by a buffer already in cache (and hence already verified) 1137bypass these checks. 1138 1139Internal Consistency Checks 1140``````````````````````````` 1141 1142After the buffer cache, the next level of metadata protection is the internal 1143record verification code built into the filesystem. 1144These checks are split between the buffer verifiers, the in-filesystem users of 1145the buffer cache, and the scrub code itself, depending on the amount of higher 1146level context required. 1147The scope of checking is still internal to the block. 1148These higher level checking functions answer these questions: 1149 1150- Does the type of data stored in the block match what scrub is expecting? 1151 1152- Does the block belong to the owning structure that asked for the read? 1153 1154- If the block contains records, do the records fit within the block? 1155 1156- If the block tracks internal free space information, is it consistent with 1157 the record areas? 1158 1159- Are the records contained inside the block free of obvious corruptions? 1160 1161Record checks in this category are more rigorous and more time-intensive. 1162For example, block pointers and inumbers are checked to ensure that they point 1163within the dynamically allocated parts of an allocation group and within 1164the filesystem. 1165Names are checked for invalid characters, and flags are checked for invalid 1166combinations. 1167Other record attributes are checked for sensible values. 1168Btree records spanning an interval of the btree keyspace are checked for 1169correct order and lack of mergeability (except for file fork mappings). 1170For performance reasons, regular code may skip some of these checks unless 1171debugging is enabled or a write is about to occur. 1172Scrub functions, of course, must check all possible problems. 1173 1174Validation of Userspace-Controlled Record Attributes 1175```````````````````````````````````````````````````` 1176 1177Various pieces of filesystem metadata are directly controlled by userspace. 1178Because of this nature, validation work cannot be more precise than checking 1179that a value is within the possible range. 1180These fields include: 1181 1182- Superblock fields controlled by mount options 1183- Filesystem labels 1184- File timestamps 1185- File permissions 1186- File size 1187- File flags 1188- Names present in directory entries, extended attribute keys, and filesystem 1189 labels 1190- Extended attribute key namespaces 1191- Extended attribute values 1192- File data block contents 1193- Quota limits 1194- Quota timer expiration (if resource usage exceeds the soft limit) 1195 1196Cross-Referencing Space Metadata 1197```````````````````````````````` 1198 1199After internal block checks, the next higher level of checking is 1200cross-referencing records between metadata structures. 1201For regular runtime code, the cost of these checks is considered to be 1202prohibitively expensive, but as scrub is dedicated to rooting out 1203inconsistencies, it must pursue all avenues of inquiry. 1204The exact set of cross-referencing is highly dependent on the context of the 1205data structure being checked. 1206 1207The XFS btree code has keyspace scanning functions that online fsck uses to 1208cross reference one structure with another. 1209Specifically, scrub can scan the key space of an index to determine if that 1210keyspace is fully, sparsely, or not at all mapped to records. 1211For the reverse mapping btree, it is possible to mask parts of the key for the 1212purposes of performing a keyspace scan so that scrub can decide if the rmap 1213btree contains records mapping a certain extent of physical space without the 1214sparsenses of the rest of the rmap keyspace getting in the way. 1215 1216Btree blocks undergo the following checks before cross-referencing: 1217 1218- Does the type of data stored in the block match what scrub is expecting? 1219 1220- Does the block belong to the owning structure that asked for the read? 1221 1222- Do the records fit within the block? 1223 1224- Are the records contained inside the block free of obvious corruptions? 1225 1226- Are the name hashes in the correct order? 1227 1228- Do node pointers within the btree point to valid block addresses for the type 1229 of btree? 1230 1231- Do child pointers point towards the leaves? 1232 1233- Do sibling pointers point across the same level? 1234 1235- For each node block record, does the record key accurate reflect the contents 1236 of the child block? 1237 1238Space allocation records are cross-referenced as follows: 1239 12401. Any space mentioned by any metadata structure are cross-referenced as 1241 follows: 1242 1243 - Does the reverse mapping index list only the appropriate owner as the 1244 owner of each block? 1245 1246 - Are none of the blocks claimed as free space? 1247 1248 - If these aren't file data blocks, are none of the blocks claimed as space 1249 shared by different owners? 1250 12512. Btree blocks are cross-referenced as follows: 1252 1253 - Everything in class 1 above. 1254 1255 - If there's a parent node block, do the keys listed for this block match the 1256 keyspace of this block? 1257 1258 - Do the sibling pointers point to valid blocks? Of the same level? 1259 1260 - Do the child pointers point to valid blocks? Of the next level down? 1261 12623. Free space btree records are cross-referenced as follows: 1263 1264 - Everything in class 1 and 2 above. 1265 1266 - Does the reverse mapping index list no owners of this space? 1267 1268 - Is this space not claimed by the inode index for inodes? 1269 1270 - Is it not mentioned by the reference count index? 1271 1272 - Is there a matching record in the other free space btree? 1273 12744. Inode btree records are cross-referenced as follows: 1275 1276 - Everything in class 1 and 2 above. 1277 1278 - Is there a matching record in free inode btree? 1279 1280 - Do cleared bits in the holemask correspond with inode clusters? 1281 1282 - Do set bits in the freemask correspond with inode records with zero link 1283 count? 1284 12855. Inode records are cross-referenced as follows: 1286 1287 - Everything in class 1. 1288 1289 - Do all the fields that summarize information about the file forks actually 1290 match those forks? 1291 1292 - Does each inode with zero link count correspond to a record in the free 1293 inode btree? 1294 12956. File fork space mapping records are cross-referenced as follows: 1296 1297 - Everything in class 1 and 2 above. 1298 1299 - Is this space not mentioned by the inode btrees? 1300 1301 - If this is a CoW fork mapping, does it correspond to a CoW entry in the 1302 reference count btree? 1303 13047. Reference count records are cross-referenced as follows: 1305 1306 - Everything in class 1 and 2 above. 1307 1308 - Within the space subkeyspace of the rmap btree (that is to say, all 1309 records mapped to a particular space extent and ignoring the owner info), 1310 are there the same number of reverse mapping records for each block as the 1311 reference count record claims? 1312 1313Proposed patchsets are the series to find gaps in 1314`refcount btree 1315<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=scrub-detect-refcount-gaps>`_, 1316`inode btree 1317<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=scrub-detect-inobt-gaps>`_, and 1318`rmap btree 1319<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=scrub-detect-rmapbt-gaps>`_ records; 1320to find 1321`mergeable records 1322<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=scrub-detect-mergeable-records>`_; 1323and to 1324`improve cross referencing with rmap 1325<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=scrub-strengthen-rmap-checking>`_ 1326before starting a repair. 1327 1328Checking Extended Attributes 1329```````````````````````````` 1330 1331Extended attributes implement a key-value store that enable fragments of data 1332to be attached to any file. 1333Both the kernel and userspace can access the keys and values, subject to 1334namespace and privilege restrictions. 1335Most typically these fragments are metadata about the file -- origins, security 1336contexts, user-supplied labels, indexing information, etc. 1337 1338Names can be as long as 255 bytes and can exist in several different 1339namespaces. 1340Values can be as large as 64KB. 1341A file's extended attributes are stored in blocks mapped by the attr fork. 1342The mappings point to leaf blocks, remote value blocks, or dabtree blocks. 1343Block 0 in the attribute fork is always the top of the structure, but otherwise 1344each of the three types of blocks can be found at any offset in the attr fork. 1345Leaf blocks contain attribute key records that point to the name and the value. 1346Names are always stored elsewhere in the same leaf block. 1347Values that are less than 3/4 the size of a filesystem block are also stored 1348elsewhere in the same leaf block. 1349Remote value blocks contain values that are too large to fit inside a leaf. 1350If the leaf information exceeds a single filesystem block, a dabtree (also 1351rooted at block 0) is created to map hashes of the attribute names to leaf 1352blocks in the attr fork. 1353 1354Checking an extended attribute structure is not so straightforward due to the 1355lack of separation between attr blocks and index blocks. 1356Scrub must read each block mapped by the attr fork and ignore the non-leaf 1357blocks: 1358 13591. Walk the dabtree in the attr fork (if present) to ensure that there are no 1360 irregularities in the blocks or dabtree mappings that do not point to 1361 attr leaf blocks. 1362 13632. Walk the blocks of the attr fork looking for leaf blocks. 1364 For each entry inside a leaf: 1365 1366 a. Validate that the name does not contain invalid characters. 1367 1368 b. Read the attr value. 1369 This performs a named lookup of the attr name to ensure the correctness 1370 of the dabtree. 1371 If the value is stored in a remote block, this also validates the 1372 integrity of the remote value block. 1373 1374Checking and Cross-Referencing Directories 1375`````````````````````````````````````````` 1376 1377The filesystem directory tree is a directed acylic graph structure, with files 1378constituting the nodes, and directory entries (dirents) constituting the edges. 1379Directories are a special type of file containing a set of mappings from a 1380255-byte sequence (name) to an inumber. 1381These are called directory entries, or dirents for short. 1382Each directory file must have exactly one directory pointing to the file. 1383A root directory points to itself. 1384Directory entries point to files of any type. 1385Each non-directory file may have multiple directories point to it. 1386 1387In XFS, directories are implemented as a file containing up to three 32GB 1388partitions. 1389The first partition contains directory entry data blocks. 1390Each data block contains variable-sized records associating a user-provided 1391name with an inumber and, optionally, a file type. 1392If the directory entry data grows beyond one block, the second partition (which 1393exists as post-EOF extents) is populated with a block containing free space 1394information and an index that maps hashes of the dirent names to directory data 1395blocks in the first partition. 1396This makes directory name lookups very fast. 1397If this second partition grows beyond one block, the third partition is 1398populated with a linear array of free space information for faster 1399expansions. 1400If the free space has been separated and the second partition grows again 1401beyond one block, then a dabtree is used to map hashes of dirent names to 1402directory data blocks. 1403 1404Checking a directory is pretty straightforward: 1405 14061. Walk the dabtree in the second partition (if present) to ensure that there 1407 are no irregularities in the blocks or dabtree mappings that do not point to 1408 dirent blocks. 1409 14102. Walk the blocks of the first partition looking for directory entries. 1411 Each dirent is checked as follows: 1412 1413 a. Does the name contain no invalid characters? 1414 1415 b. Does the inumber correspond to an actual, allocated inode? 1416 1417 c. Does the child inode have a nonzero link count? 1418 1419 d. If a file type is included in the dirent, does it match the type of the 1420 inode? 1421 1422 e. If the child is a subdirectory, does the child's dotdot pointer point 1423 back to the parent? 1424 1425 f. If the directory has a second partition, perform a named lookup of the 1426 dirent name to ensure the correctness of the dabtree. 1427 14283. Walk the free space list in the third partition (if present) to ensure that 1429 the free spaces it describes are really unused. 1430 1431Checking operations involving :ref:`parents <dirparent>` and 1432:ref:`file link counts <nlinks>` are discussed in more detail in later 1433sections. 1434 1435Checking Directory/Attribute Btrees 1436^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1437 1438As stated in previous sections, the directory/attribute btree (dabtree) index 1439maps user-provided names to improve lookup times by avoiding linear scans. 1440Internally, it maps a 32-bit hash of the name to a block offset within the 1441appropriate file fork. 1442 1443The internal structure of a dabtree closely resembles the btrees that record 1444fixed-size metadata records -- each dabtree block contains a magic number, a 1445checksum, sibling pointers, a UUID, a tree level, and a log sequence number. 1446The format of leaf and node records are the same -- each entry points to the 1447next level down in the hierarchy, with dabtree node records pointing to dabtree 1448leaf blocks, and dabtree leaf records pointing to non-dabtree blocks elsewhere 1449in the fork. 1450 1451Checking and cross-referencing the dabtree is very similar to what is done for 1452space btrees: 1453 1454- Does the type of data stored in the block match what scrub is expecting? 1455 1456- Does the block belong to the owning structure that asked for the read? 1457 1458- Do the records fit within the block? 1459 1460- Are the records contained inside the block free of obvious corruptions? 1461 1462- Are the name hashes in the correct order? 1463 1464- Do node pointers within the dabtree point to valid fork offsets for dabtree 1465 blocks? 1466 1467- Do leaf pointers within the dabtree point to valid fork offsets for directory 1468 or attr leaf blocks? 1469 1470- Do child pointers point towards the leaves? 1471 1472- Do sibling pointers point across the same level? 1473 1474- For each dabtree node record, does the record key accurate reflect the 1475 contents of the child dabtree block? 1476 1477- For each dabtree leaf record, does the record key accurate reflect the 1478 contents of the directory or attr block? 1479 1480Cross-Referencing Summary Counters 1481`````````````````````````````````` 1482 1483XFS maintains three classes of summary counters: available resources, quota 1484resource usage, and file link counts. 1485 1486In theory, the amount of available resources (data blocks, inodes, realtime 1487extents) can be found by walking the entire filesystem. 1488This would make for very slow reporting, so a transactional filesystem can 1489maintain summaries of this information in the superblock. 1490Cross-referencing these values against the filesystem metadata should be a 1491simple matter of walking the free space and inode metadata in each AG and the 1492realtime bitmap, but there are complications that will be discussed in 1493:ref:`more detail <fscounters>` later. 1494 1495:ref:`Quota usage <quotacheck>` and :ref:`file link count <nlinks>` 1496checking are sufficiently complicated to warrant separate sections. 1497 1498Post-Repair Reverification 1499`````````````````````````` 1500 1501After performing a repair, the checking code is run a second time to validate 1502the new structure, and the results of the health assessment are recorded 1503internally and returned to the calling process. 1504This step is critical for enabling system administrator to monitor the status 1505of the filesystem and the progress of any repairs. 1506For developers, it is a useful means to judge the efficacy of error detection 1507and correction in the online and offline checking tools. 1508 1509Eventual Consistency vs. Online Fsck 1510------------------------------------ 1511 1512Complex operations can make modifications to multiple per-AG data structures 1513with a chain of transactions. 1514These chains, once committed to the log, are restarted during log recovery if 1515the system crashes while processing the chain. 1516Because the AG header buffers are unlocked between transactions within a chain, 1517online checking must coordinate with chained operations that are in progress to 1518avoid incorrectly detecting inconsistencies due to pending chains. 1519Furthermore, online repair must not run when operations are pending because 1520the metadata are temporarily inconsistent with each other, and rebuilding is 1521not possible. 1522 1523Only online fsck has this requirement of total consistency of AG metadata, and 1524should be relatively rare as compared to filesystem change operations. 1525Online fsck coordinates with transaction chains as follows: 1526 1527* For each AG, maintain a count of intent items targeting that AG. 1528 The count should be bumped whenever a new item is added to the chain. 1529 The count should be dropped when the filesystem has locked the AG header 1530 buffers and finished the work. 1531 1532* When online fsck wants to examine an AG, it should lock the AG header 1533 buffers to quiesce all transaction chains that want to modify that AG. 1534 If the count is zero, proceed with the checking operation. 1535 If it is nonzero, cycle the buffer locks to allow the chain to make forward 1536 progress. 1537 1538This may lead to online fsck taking a long time to complete, but regular 1539filesystem updates take precedence over background checking activity. 1540Details about the discovery of this situation are presented in the 1541:ref:`next section <chain_coordination>`, and details about the solution 1542are presented :ref:`after that<intent_drains>`. 1543 1544.. _chain_coordination: 1545 1546Discovery of the Problem 1547```````````````````````` 1548 1549Midway through the development of online scrubbing, the fsstress tests 1550uncovered a misinteraction between online fsck and compound transaction chains 1551created by other writer threads that resulted in false reports of metadata 1552inconsistency. 1553The root cause of these reports is the eventual consistency model introduced by 1554the expansion of deferred work items and compound transaction chains when 1555reverse mapping and reflink were introduced. 1556 1557Originally, transaction chains were added to XFS to avoid deadlocks when 1558unmapping space from files. 1559Deadlock avoidance rules require that AGs only be locked in increasing order, 1560which makes it impossible (say) to use a single transaction to free a space 1561extent in AG 7 and then try to free a now superfluous block mapping btree block 1562in AG 3. 1563To avoid these kinds of deadlocks, XFS creates Extent Freeing Intent (EFI) log 1564items to commit to freeing some space in one transaction while deferring the 1565actual metadata updates to a fresh transaction. 1566The transaction sequence looks like this: 1567 15681. The first transaction contains a physical update to the file's block mapping 1569 structures to remove the mapping from the btree blocks. 1570 It then attaches to the in-memory transaction an action item to schedule 1571 deferred freeing of space. 1572 Concretely, each transaction maintains a list of ``struct 1573 xfs_defer_pending`` objects, each of which maintains a list of ``struct 1574 xfs_extent_free_item`` objects. 1575 Returning to the example above, the action item tracks the freeing of both 1576 the unmapped space from AG 7 and the block mapping btree (BMBT) block from 1577 AG 3. 1578 Deferred frees recorded in this manner are committed in the log by creating 1579 an EFI log item from the ``struct xfs_extent_free_item`` object and 1580 attaching the log item to the transaction. 1581 When the log is persisted to disk, the EFI item is written into the ondisk 1582 transaction record. 1583 EFIs can list up to 16 extents to free, all sorted in AG order. 1584 15852. The second transaction contains a physical update to the free space btrees 1586 of AG 3 to release the former BMBT block and a second physical update to the 1587 free space btrees of AG 7 to release the unmapped file space. 1588 Observe that the physical updates are resequenced in the correct order 1589 when possible. 1590 Attached to the transaction is a an extent free done (EFD) log item. 1591 The EFD contains a pointer to the EFI logged in transaction #1 so that log 1592 recovery can tell if the EFI needs to be replayed. 1593 1594If the system goes down after transaction #1 is written back to the filesystem 1595but before #2 is committed, a scan of the filesystem metadata would show 1596inconsistent filesystem metadata because there would not appear to be any owner 1597of the unmapped space. 1598Happily, log recovery corrects this inconsistency for us -- when recovery finds 1599an intent log item but does not find a corresponding intent done item, it will 1600reconstruct the incore state of the intent item and finish it. 1601In the example above, the log must replay both frees described in the recovered 1602EFI to complete the recovery phase. 1603 1604There are subtleties to XFS' transaction chaining strategy to consider: 1605 1606* Log items must be added to a transaction in the correct order to prevent 1607 conflicts with principal objects that are not held by the transaction. 1608 In other words, all per-AG metadata updates for an unmapped block must be 1609 completed before the last update to free the extent, and extents should not 1610 be reallocated until that last update commits to the log. 1611 1612* AG header buffers are released between each transaction in a chain. 1613 This means that other threads can observe an AG in an intermediate state, 1614 but as long as the first subtlety is handled, this should not affect the 1615 correctness of filesystem operations. 1616 1617* Unmounting the filesystem flushes all pending work to disk, which means that 1618 offline fsck never sees the temporary inconsistencies caused by deferred 1619 work item processing. 1620 1621In this manner, XFS employs a form of eventual consistency to avoid deadlocks 1622and increase parallelism. 1623 1624During the design phase of the reverse mapping and reflink features, it was 1625decided that it was impractical to cram all the reverse mapping updates for a 1626single filesystem change into a single transaction because a single file 1627mapping operation can explode into many small updates: 1628 1629* The block mapping update itself 1630* A reverse mapping update for the block mapping update 1631* Fixing the freelist 1632* A reverse mapping update for the freelist fix 1633 1634* A shape change to the block mapping btree 1635* A reverse mapping update for the btree update 1636* Fixing the freelist (again) 1637* A reverse mapping update for the freelist fix 1638 1639* An update to the reference counting information 1640* A reverse mapping update for the refcount update 1641* Fixing the freelist (a third time) 1642* A reverse mapping update for the freelist fix 1643 1644* Freeing any space that was unmapped and not owned by any other file 1645* Fixing the freelist (a fourth time) 1646* A reverse mapping update for the freelist fix 1647 1648* Freeing the space used by the block mapping btree 1649* Fixing the freelist (a fifth time) 1650* A reverse mapping update for the freelist fix 1651 1652Free list fixups are not usually needed more than once per AG per transaction 1653chain, but it is theoretically possible if space is very tight. 1654For copy-on-write updates this is even worse, because this must be done once to 1655remove the space from a staging area and again to map it into the file! 1656 1657To deal with this explosion in a calm manner, XFS expands its use of deferred 1658work items to cover most reverse mapping updates and all refcount updates. 1659This reduces the worst case size of transaction reservations by breaking the 1660work into a long chain of small updates, which increases the degree of eventual 1661consistency in the system. 1662Again, this generally isn't a problem because XFS orders its deferred work 1663items carefully to avoid resource reuse conflicts between unsuspecting threads. 1664 1665However, online fsck changes the rules -- remember that although physical 1666updates to per-AG structures are coordinated by locking the buffers for AG 1667headers, buffer locks are dropped between transactions. 1668Once scrub acquires resources and takes locks for a data structure, it must do 1669all the validation work without releasing the lock. 1670If the main lock for a space btree is an AG header buffer lock, scrub may have 1671interrupted another thread that is midway through finishing a chain. 1672For example, if a thread performing a copy-on-write has completed a reverse 1673mapping update but not the corresponding refcount update, the two AG btrees 1674will appear inconsistent to scrub and an observation of corruption will be 1675recorded. This observation will not be correct. 1676If a repair is attempted in this state, the results will be catastrophic! 1677 1678Several other solutions to this problem were evaluated upon discovery of this 1679flaw and rejected: 1680 16811. Add a higher level lock to allocation groups and require writer threads to 1682 acquire the higher level lock in AG order before making any changes. 1683 This would be very difficult to implement in practice because it is 1684 difficult to determine which locks need to be obtained, and in what order, 1685 without simulating the entire operation. 1686 Performing a dry run of a file operation to discover necessary locks would 1687 make the filesystem very slow. 1688 16892. Make the deferred work coordinator code aware of consecutive intent items 1690 targeting the same AG and have it hold the AG header buffers locked across 1691 the transaction roll between updates. 1692 This would introduce a lot of complexity into the coordinator since it is 1693 only loosely coupled with the actual deferred work items. 1694 It would also fail to solve the problem because deferred work items can 1695 generate new deferred subtasks, but all subtasks must be complete before 1696 work can start on a new sibling task. 1697 16983. Teach online fsck to walk all transactions waiting for whichever lock(s) 1699 protect the data structure being scrubbed to look for pending operations. 1700 The checking and repair operations must factor these pending operations into 1701 the evaluations being performed. 1702 This solution is a nonstarter because it is *extremely* invasive to the main 1703 filesystem. 1704 1705.. _intent_drains: 1706 1707Intent Drains 1708````````````` 1709 1710Online fsck uses an atomic intent item counter and lock cycling to coordinate 1711with transaction chains. 1712There are two key properties to the drain mechanism. 1713First, the counter is incremented when a deferred work item is *queued* to a 1714transaction, and it is decremented after the associated intent done log item is 1715*committed* to another transaction. 1716The second property is that deferred work can be added to a transaction without 1717holding an AG header lock, but per-AG work items cannot be marked done without 1718locking that AG header buffer to log the physical updates and the intent done 1719log item. 1720The first property enables scrub to yield to running transaction chains, which 1721is an explicit deprioritization of online fsck to benefit file operations. 1722The second property of the drain is key to the correct coordination of scrub, 1723since scrub will always be able to decide if a conflict is possible. 1724 1725For regular filesystem code, the drain works as follows: 1726 17271. Call the appropriate subsystem function to add a deferred work item to a 1728 transaction. 1729 17302. The function calls ``xfs_defer_drain_bump`` to increase the counter. 1731 17323. When the deferred item manager wants to finish the deferred work item, it 1733 calls ``->finish_item`` to complete it. 1734 17354. The ``->finish_item`` implementation logs some changes and calls 1736 ``xfs_defer_drain_drop`` to decrease the sloppy counter and wake up any threads 1737 waiting on the drain. 1738 17395. The subtransaction commits, which unlocks the resource associated with the 1740 intent item. 1741 1742For scrub, the drain works as follows: 1743 17441. Lock the resource(s) associated with the metadata being scrubbed. 1745 For example, a scan of the refcount btree would lock the AGI and AGF header 1746 buffers. 1747 17482. If the counter is zero (``xfs_defer_drain_busy`` returns false), there are no 1749 chains in progress and the operation may proceed. 1750 17513. Otherwise, release the resources grabbed in step 1. 1752 17534. Wait for the intent counter to reach zero (``xfs_defer_drain_intents``), then go 1754 back to step 1 unless a signal has been caught. 1755 1756To avoid polling in step 4, the drain provides a waitqueue for scrub threads to 1757be woken up whenever the intent count drops to zero. 1758 1759The proposed patchset is the 1760`scrub intent drain series 1761<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=scrub-drain-intents>`_. 1762 1763.. _jump_labels: 1764 1765Static Keys (aka Jump Label Patching) 1766````````````````````````````````````` 1767 1768Online fsck for XFS separates the regular filesystem from the checking and 1769repair code as much as possible. 1770However, there are a few parts of online fsck (such as the intent drains, and 1771later, live update hooks) where it is useful for the online fsck code to know 1772what's going on in the rest of the filesystem. 1773Since it is not expected that online fsck will be constantly running in the 1774background, it is very important to minimize the runtime overhead imposed by 1775these hooks when online fsck is compiled into the kernel but not actively 1776running on behalf of userspace. 1777Taking locks in the hot path of a writer thread to access a data structure only 1778to find that no further action is necessary is expensive -- on the author's 1779computer, this have an overhead of 40-50ns per access. 1780Fortunately, the kernel supports dynamic code patching, which enables XFS to 1781replace a static branch to hook code with ``nop`` sleds when online fsck isn't 1782running. 1783This sled has an overhead of however long it takes the instruction decoder to 1784skip past the sled, which seems to be on the order of less than 1ns and 1785does not access memory outside of instruction fetching. 1786 1787When online fsck enables the static key, the sled is replaced with an 1788unconditional branch to call the hook code. 1789The switchover is quite expensive (~22000ns) but is paid entirely by the 1790program that invoked online fsck, and can be amortized if multiple threads 1791enter online fsck at the same time, or if multiple filesystems are being 1792checked at the same time. 1793Changing the branch direction requires taking the CPU hotplug lock, and since 1794CPU initialization requires memory allocation, online fsck must be careful not 1795to change a static key while holding any locks or resources that could be 1796accessed in the memory reclaim paths. 1797To minimize contention on the CPU hotplug lock, care should be taken not to 1798enable or disable static keys unnecessarily. 1799 1800Because static keys are intended to minimize hook overhead for regular 1801filesystem operations when xfs_scrub is not running, the intended usage 1802patterns are as follows: 1803 1804- The hooked part of XFS should declare a static-scoped static key that 1805 defaults to false. 1806 The ``DEFINE_STATIC_KEY_FALSE`` macro takes care of this. 1807 The static key itself should be declared as a ``static`` variable. 1808 1809- When deciding to invoke code that's only used by scrub, the regular 1810 filesystem should call the ``static_branch_unlikely`` predicate to avoid the 1811 scrub-only hook code if the static key is not enabled. 1812 1813- The regular filesystem should export helper functions that call 1814 ``static_branch_inc`` to enable and ``static_branch_dec`` to disable the 1815 static key. 1816 Wrapper functions make it easy to compile out the relevant code if the kernel 1817 distributor turns off online fsck at build time. 1818 1819- Scrub functions wanting to turn on scrub-only XFS functionality should call 1820 the ``xchk_fsgates_enable`` from the setup function to enable a specific 1821 hook. 1822 This must be done before obtaining any resources that are used by memory 1823 reclaim. 1824 Callers had better be sure they really need the functionality gated by the 1825 static key; the ``TRY_HARDER`` flag is useful here. 1826 1827Online scrub has resource acquisition helpers (e.g. ``xchk_perag_lock``) to 1828handle locking AGI and AGF buffers for all scrubber functions. 1829If it detects a conflict between scrub and the running transactions, it will 1830try to wait for intents to complete. 1831If the caller of the helper has not enabled the static key, the helper will 1832return -EDEADLOCK, which should result in the scrub being restarted with the 1833``TRY_HARDER`` flag set. 1834The scrub setup function should detect that flag, enable the static key, and 1835try the scrub again. 1836Scrub teardown disables all static keys obtained by ``xchk_fsgates_enable``. 1837 1838For more information, please see the kernel documentation of 1839Documentation/staging/static-keys.rst. 1840 1841.. _xfile: 1842 1843Pageable Kernel Memory 1844---------------------- 1845 1846Some online checking functions work by scanning the filesystem to build a 1847shadow copy of an ondisk metadata structure in memory and comparing the two 1848copies. 1849For online repair to rebuild a metadata structure, it must compute the record 1850set that will be stored in the new structure before it can persist that new 1851structure to disk. 1852Ideally, repairs complete with a single atomic commit that introduces 1853a new data structure. 1854To meet these goals, the kernel needs to collect a large amount of information 1855in a place that doesn't require the correct operation of the filesystem. 1856 1857Kernel memory isn't suitable because: 1858 1859* Allocating a contiguous region of memory to create a C array is very 1860 difficult, especially on 32-bit systems. 1861 1862* Linked lists of records introduce double pointer overhead which is very high 1863 and eliminate the possibility of indexed lookups. 1864 1865* Kernel memory is pinned, which can drive the system into OOM conditions. 1866 1867* The system might not have sufficient memory to stage all the information. 1868 1869At any given time, online fsck does not need to keep the entire record set in 1870memory, which means that individual records can be paged out if necessary. 1871Continued development of online fsck demonstrated that the ability to perform 1872indexed data storage would also be very useful. 1873Fortunately, the Linux kernel already has a facility for byte-addressable and 1874pageable storage: tmpfs. 1875In-kernel graphics drivers (most notably i915) take advantage of tmpfs files 1876to store intermediate data that doesn't need to be in memory at all times, so 1877that usage precedent is already established. 1878Hence, the ``xfile`` was born! 1879 1880+--------------------------------------------------------------------------+ 1881| **Historical Sidebar**: | 1882+--------------------------------------------------------------------------+ 1883| The first edition of online repair inserted records into a new btree as | 1884| it found them, which failed because filesystem could shut down with a | 1885| built data structure, which would be live after recovery finished. | 1886| | 1887| The second edition solved the half-rebuilt structure problem by storing | 1888| everything in memory, but frequently ran the system out of memory. | 1889| | 1890| The third edition solved the OOM problem by using linked lists, but the | 1891| memory overhead of the list pointers was extreme. | 1892+--------------------------------------------------------------------------+ 1893 1894xfile Access Models 1895``````````````````` 1896 1897A survey of the intended uses of xfiles suggested these use cases: 1898 18991. Arrays of fixed-sized records (space management btrees, directory and 1900 extended attribute entries) 1901 19022. Sparse arrays of fixed-sized records (quotas and link counts) 1903 19043. Large binary objects (BLOBs) of variable sizes (directory and extended 1905 attribute names and values) 1906 19074. Staging btrees in memory (reverse mapping btrees) 1908 19095. Arbitrary contents (realtime space management) 1910 1911To support the first four use cases, high level data structures wrap the xfile 1912to share functionality between online fsck functions. 1913The rest of this section discusses the interfaces that the xfile presents to 1914four of those five higher level data structures. 1915The fifth use case is discussed in the :ref:`realtime summary <rtsummary>` case 1916study. 1917 1918XFS is very record-based, which suggests that the ability to load and store 1919complete records is important. 1920To support these cases, a pair of ``xfile_load`` and ``xfile_store`` 1921functions are provided to read and persist objects into an xfile that treat any 1922error as an out of memory error. For online repair, squashing error conditions 1923in this manner is an acceptable behavior because the only reaction is to abort 1924the operation back to userspace. 1925 1926However, no discussion of file access idioms is complete without answering the 1927question, "But what about mmap?" 1928It is convenient to access storage directly with pointers, just like userspace 1929code does with regular memory. 1930Online fsck must not drive the system into OOM conditions, which means that 1931xfiles must be responsive to memory reclamation. 1932tmpfs can only push a pagecache folio to the swap cache if the folio is neither 1933pinned nor locked, which means the xfile must not pin too many folios. 1934 1935Short term direct access to xfile contents is done by locking the pagecache 1936folio and mapping it into kernel address space. Object load and store uses this 1937mechanism. Folio locks are not supposed to be held for long periods of time, so 1938long term direct access to xfile contents is done by bumping the folio refcount, 1939mapping it into kernel address space, and dropping the folio lock. 1940These long term users *must* be responsive to memory reclaim by hooking into 1941the shrinker infrastructure to know when to release folios. 1942 1943The ``xfile_get_folio`` and ``xfile_put_folio`` functions are provided to 1944retrieve the (locked) folio that backs part of an xfile and to release it. 1945The only code to use these folio lease functions are the xfarray 1946:ref:`sorting<xfarray_sort>` algorithms and the :ref:`in-memory 1947btrees<xfbtree>`. 1948 1949xfile Access Coordination 1950````````````````````````` 1951 1952For security reasons, xfiles must be owned privately by the kernel. 1953They are marked ``S_PRIVATE`` to prevent interference from the security system, 1954must never be mapped into process file descriptor tables, and their pages must 1955never be mapped into userspace processes. 1956 1957To avoid locking recursion issues with the VFS, all accesses to the shmfs file 1958are performed by manipulating the page cache directly. 1959xfile writers call the ``->write_begin`` and ``->write_end`` functions of the 1960xfile's address space to grab writable pages, copy the caller's buffer into the 1961page, and release the pages. 1962xfile readers call ``shmem_read_mapping_page_gfp`` to grab pages directly 1963before copying the contents into the caller's buffer. 1964In other words, xfiles ignore the VFS read and write code paths to avoid 1965having to create a dummy ``struct kiocb`` and to avoid taking inode and 1966freeze locks. 1967tmpfs cannot be frozen, and xfiles must not be exposed to userspace. 1968 1969If an xfile is shared between threads to stage repairs, the caller must provide 1970its own locks to coordinate access. 1971For example, if a scrub function stores scan results in an xfile and needs 1972other threads to provide updates to the scanned data, the scrub function must 1973provide a lock for all threads to share. 1974 1975.. _xfarray: 1976 1977Arrays of Fixed-Sized Records 1978````````````````````````````` 1979 1980In XFS, each type of indexed space metadata (free space, inodes, reference 1981counts, file fork space, and reverse mappings) consists of a set of fixed-size 1982records indexed with a classic B+ tree. 1983Directories have a set of fixed-size dirent records that point to the names, 1984and extended attributes have a set of fixed-size attribute keys that point to 1985names and values. 1986Quota counters and file link counters index records with numbers. 1987During a repair, scrub needs to stage new records during the gathering step and 1988retrieve them during the btree building step. 1989 1990Although this requirement can be satisfied by calling the read and write 1991methods of the xfile directly, it is simpler for callers for there to be a 1992higher level abstraction to take care of computing array offsets, to provide 1993iterator functions, and to deal with sparse records and sorting. 1994The ``xfarray`` abstraction presents a linear array for fixed-size records atop 1995the byte-accessible xfile. 1996 1997.. _xfarray_access_patterns: 1998 1999Array Access Patterns 2000^^^^^^^^^^^^^^^^^^^^^ 2001 2002Array access patterns in online fsck tend to fall into three categories. 2003Iteration of records is assumed to be necessary for all cases and will be 2004covered in the next section. 2005 2006The first type of caller handles records that are indexed by position. 2007Gaps may exist between records, and a record may be updated multiple times 2008during the collection step. 2009In other words, these callers want a sparse linearly addressed table file. 2010The typical use case are quota records or file link count records. 2011Access to array elements is performed programmatically via ``xfarray_load`` and 2012``xfarray_store`` functions, which wrap the similarly-named xfile functions to 2013provide loading and storing of array elements at arbitrary array indices. 2014Gaps are defined to be null records, and null records are defined to be a 2015sequence of all zero bytes. 2016Null records are detected by calling ``xfarray_element_is_null``. 2017They are created either by calling ``xfarray_unset`` to null out an existing 2018record or by never storing anything to an array index. 2019 2020The second type of caller handles records that are not indexed by position 2021and do not require multiple updates to a record. 2022The typical use case here is rebuilding space btrees and key/value btrees. 2023These callers can add records to the array without caring about array indices 2024via the ``xfarray_append`` function, which stores a record at the end of the 2025array. 2026For callers that require records to be presentable in a specific order (e.g. 2027rebuilding btree data), the ``xfarray_sort`` function can arrange the sorted 2028records; this function will be covered later. 2029 2030The third type of caller is a bag, which is useful for counting records. 2031The typical use case here is constructing space extent reference counts from 2032reverse mapping information. 2033Records can be put in the bag in any order, they can be removed from the bag 2034at any time, and uniqueness of records is left to callers. 2035The ``xfarray_store_anywhere`` function is used to insert a record in any 2036null record slot in the bag; and the ``xfarray_unset`` function removes a 2037record from the bag. 2038 2039The proposed patchset is the 2040`big in-memory array 2041<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=big-array>`_. 2042 2043Iterating Array Elements 2044^^^^^^^^^^^^^^^^^^^^^^^^ 2045 2046Most users of the xfarray require the ability to iterate the records stored in 2047the array. 2048Callers can probe every possible array index with the following: 2049 2050.. code-block:: c 2051 2052 xfarray_idx_t i; 2053 foreach_xfarray_idx(array, i) { 2054 xfarray_load(array, i, &rec); 2055 2056 /* do something with rec */ 2057 } 2058 2059All users of this idiom must be prepared to handle null records or must already 2060know that there aren't any. 2061 2062For xfarray users that want to iterate a sparse array, the ``xfarray_iter`` 2063function ignores indices in the xfarray that have never been written to by 2064calling ``xfile_seek_data`` (which internally uses ``SEEK_DATA``) to skip areas 2065of the array that are not populated with memory pages. 2066Once it finds a page, it will skip the zeroed areas of the page. 2067 2068.. code-block:: c 2069 2070 xfarray_idx_t i = XFARRAY_CURSOR_INIT; 2071 while ((ret = xfarray_iter(array, &i, &rec)) == 1) { 2072 /* do something with rec */ 2073 } 2074 2075.. _xfarray_sort: 2076 2077Sorting Array Elements 2078^^^^^^^^^^^^^^^^^^^^^^ 2079 2080During the fourth demonstration of online repair, a community reviewer remarked 2081that for performance reasons, online repair ought to load batches of records 2082into btree record blocks instead of inserting records into a new btree one at a 2083time. 2084The btree insertion code in XFS is responsible for maintaining correct ordering 2085of the records, so naturally the xfarray must also support sorting the record 2086set prior to bulk loading. 2087 2088Case Study: Sorting xfarrays 2089~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 2090 2091The sorting algorithm used in the xfarray is actually a combination of adaptive 2092quicksort and a heapsort subalgorithm in the spirit of 2093`Sedgewick <https://algs4.cs.princeton.edu/23quicksort/>`_ and 2094`pdqsort <https://github.com/orlp/pdqsort>`_, with customizations for the Linux 2095kernel. 2096To sort records in a reasonably short amount of time, ``xfarray`` takes 2097advantage of the binary subpartitioning offered by quicksort, but it also uses 2098heapsort to hedge against performance collapse if the chosen quicksort pivots 2099are poor. 2100Both algorithms are (in general) O(n * lg(n)), but there is a wide performance 2101gulf between the two implementations. 2102 2103The Linux kernel already contains a reasonably fast implementation of heapsort. 2104It only operates on regular C arrays, which limits the scope of its usefulness. 2105There are two key places where the xfarray uses it: 2106 2107* Sorting any record subset backed by a single xfile page. 2108 2109* Loading a small number of xfarray records from potentially disparate parts 2110 of the xfarray into a memory buffer, and sorting the buffer. 2111 2112In other words, ``xfarray`` uses heapsort to constrain the nested recursion of 2113quicksort, thereby mitigating quicksort's worst runtime behavior. 2114 2115Choosing a quicksort pivot is a tricky business. 2116A good pivot splits the set to sort in half, leading to the divide and conquer 2117behavior that is crucial to O(n * lg(n)) performance. 2118A poor pivot barely splits the subset at all, leading to O(n\ :sup:`2`) 2119runtime. 2120The xfarray sort routine tries to avoid picking a bad pivot by sampling nine 2121records into a memory buffer and using the kernel heapsort to identify the 2122median of the nine. 2123 2124Most modern quicksort implementations employ Tukey's "ninther" to select a 2125pivot from a classic C array. 2126Typical ninther implementations pick three unique triads of records, sort each 2127of the triads, and then sort the middle value of each triad to determine the 2128ninther value. 2129As stated previously, however, xfile accesses are not entirely cheap. 2130It turned out to be much more performant to read the nine elements into a 2131memory buffer, run the kernel's in-memory heapsort on the buffer, and choose 2132the 4th element of that buffer as the pivot. 2133Tukey's ninthers are described in J. W. Tukey, `The ninther, a technique for 2134low-effort robust (resistant) location in large samples`, in *Contributions to 2135Survey Sampling and Applied Statistics*, edited by H. David, (Academic Press, 21361978), pp. 251–257. 2137 2138The partitioning of quicksort is fairly textbook -- rearrange the record 2139subset around the pivot, then set up the current and next stack frames to 2140sort with the larger and the smaller halves of the pivot, respectively. 2141This keeps the stack space requirements to log2(record count). 2142 2143As a final performance optimization, the hi and lo scanning phase of quicksort 2144keeps examined xfile pages mapped in the kernel for as long as possible to 2145reduce map/unmap cycles. 2146Surprisingly, this reduces overall sort runtime by nearly half again after 2147accounting for the application of heapsort directly onto xfile pages. 2148 2149.. _xfblob: 2150 2151Blob Storage 2152```````````` 2153 2154Extended attributes and directories add an additional requirement for staging 2155records: arbitrary byte sequences of finite length. 2156Each directory entry record needs to store entry name, 2157and each extended attribute needs to store both the attribute name and value. 2158The names, keys, and values can consume a large amount of memory, so the 2159``xfblob`` abstraction was created to simplify management of these blobs 2160atop an xfile. 2161 2162Blob arrays provide ``xfblob_load`` and ``xfblob_store`` functions to retrieve 2163and persist objects. 2164The store function returns a magic cookie for every object that it persists. 2165Later, callers provide this cookie to the ``xblob_load`` to recall the object. 2166The ``xfblob_free`` function frees a specific blob, and the ``xfblob_truncate`` 2167function frees them all because compaction is not needed. 2168 2169The details of repairing directories and extended attributes will be discussed 2170in a subsequent section about atomic extent swapping. 2171However, it should be noted that these repair functions only use blob storage 2172to cache a small number of entries before adding them to a temporary ondisk 2173file, which is why compaction is not required. 2174 2175The proposed patchset is at the start of the 2176`extended attribute repair 2177<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-xattrs>`_ series. 2178 2179.. _xfbtree: 2180 2181In-Memory B+Trees 2182````````````````` 2183 2184The chapter about :ref:`secondary metadata<secondary_metadata>` mentioned that 2185checking and repairing of secondary metadata commonly requires coordination 2186between a live metadata scan of the filesystem and writer threads that are 2187updating that metadata. 2188Keeping the scan data up to date requires requires the ability to propagate 2189metadata updates from the filesystem into the data being collected by the scan. 2190This *can* be done by appending concurrent updates into a separate log file and 2191applying them before writing the new metadata to disk, but this leads to 2192unbounded memory consumption if the rest of the system is very busy. 2193Another option is to skip the side-log and commit live updates from the 2194filesystem directly into the scan data, which trades more overhead for a lower 2195maximum memory requirement. 2196In both cases, the data structure holding the scan results must support indexed 2197access to perform well. 2198 2199Given that indexed lookups of scan data is required for both strategies, online 2200fsck employs the second strategy of committing live updates directly into 2201scan data. 2202Because xfarrays are not indexed and do not enforce record ordering, they 2203are not suitable for this task. 2204Conveniently, however, XFS has a library to create and maintain ordered reverse 2205mapping records: the existing rmap btree code! 2206If only there was a means to create one in memory. 2207 2208Recall that the :ref:`xfile <xfile>` abstraction represents memory pages as a 2209regular file, which means that the kernel can create byte or block addressable 2210virtual address spaces at will. 2211The XFS buffer cache specializes in abstracting IO to block-oriented address 2212spaces, which means that adaptation of the buffer cache to interface with 2213xfiles enables reuse of the entire btree library. 2214Btrees built atop an xfile are collectively known as ``xfbtrees``. 2215The next few sections describe how they actually work. 2216 2217The proposed patchset is the 2218`in-memory btree 2219<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=in-memory-btrees>`_ 2220series. 2221 2222Using xfiles as a Buffer Cache Target 2223^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2224 2225Two modifications are necessary to support xfiles as a buffer cache target. 2226The first is to make it possible for the ``struct xfs_buftarg`` structure to 2227host the ``struct xfs_buf`` rhashtable, because normally those are held by a 2228per-AG structure. 2229The second change is to modify the buffer ``ioapply`` function to "read" cached 2230pages from the xfile and "write" cached pages back to the xfile. 2231Multiple access to individual buffers is controlled by the ``xfs_buf`` lock, 2232since the xfile does not provide any locking on its own. 2233With this adaptation in place, users of the xfile-backed buffer cache use 2234exactly the same APIs as users of the disk-backed buffer cache. 2235The separation between xfile and buffer cache implies higher memory usage since 2236they do not share pages, but this property could some day enable transactional 2237updates to an in-memory btree. 2238Today, however, it simply eliminates the need for new code. 2239 2240Space Management with an xfbtree 2241^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2242 2243Space management for an xfile is very simple -- each btree block is one memory 2244page in size. 2245These blocks use the same header format as an on-disk btree, but the in-memory 2246block verifiers ignore the checksums, assuming that xfile memory is no more 2247corruption-prone than regular DRAM. 2248Reusing existing code here is more important than absolute memory efficiency. 2249 2250The very first block of an xfile backing an xfbtree contains a header block. 2251The header describes the owner, height, and the block number of the root 2252xfbtree block. 2253 2254To allocate a btree block, use ``xfile_seek_data`` to find a gap in the file. 2255If there are no gaps, create one by extending the length of the xfile. 2256Preallocate space for the block with ``xfile_prealloc``, and hand back the 2257location. 2258To free an xfbtree block, use ``xfile_discard`` (which internally uses 2259``FALLOC_FL_PUNCH_HOLE``) to remove the memory page from the xfile. 2260 2261Populating an xfbtree 2262^^^^^^^^^^^^^^^^^^^^^ 2263 2264An online fsck function that wants to create an xfbtree should proceed as 2265follows: 2266 22671. Call ``xfile_create`` to create an xfile. 2268 22692. Call ``xfs_alloc_memory_buftarg`` to create a buffer cache target structure 2270 pointing to the xfile. 2271 22723. Pass the buffer cache target, buffer ops, and other information to 2273 ``xfbtree_init`` to initialize the passed in ``struct xfbtree`` and write an 2274 initial root block to the xfile. 2275 Each btree type should define a wrapper that passes necessary arguments to 2276 the creation function. 2277 For example, rmap btrees define ``xfs_rmapbt_mem_create`` to take care of 2278 all the necessary details for callers. 2279 22804. Pass the xfbtree object to the btree cursor creation function for the 2281 btree type. 2282 Following the example above, ``xfs_rmapbt_mem_cursor`` takes care of this 2283 for callers. 2284 22855. Pass the btree cursor to the regular btree functions to make queries against 2286 and to update the in-memory btree. 2287 For example, a btree cursor for an rmap xfbtree can be passed to the 2288 ``xfs_rmap_*`` functions just like any other btree cursor. 2289 See the :ref:`next section<xfbtree_commit>` for information on dealing with 2290 xfbtree updates that are logged to a transaction. 2291 22926. When finished, delete the btree cursor, destroy the xfbtree object, free the 2293 buffer target, and the destroy the xfile to release all resources. 2294 2295.. _xfbtree_commit: 2296 2297Committing Logged xfbtree Buffers 2298^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2299 2300Although it is a clever hack to reuse the rmap btree code to handle the staging 2301structure, the ephemeral nature of the in-memory btree block storage presents 2302some challenges of its own. 2303The XFS transaction manager must not commit buffer log items for buffers backed 2304by an xfile because the log format does not understand updates for devices 2305other than the data device. 2306An ephemeral xfbtree probably will not exist by the time the AIL checkpoints 2307log transactions back into the filesystem, and certainly won't exist during 2308log recovery. 2309For these reasons, any code updating an xfbtree in transaction context must 2310remove the buffer log items from the transaction and write the updates into the 2311backing xfile before committing or cancelling the transaction. 2312 2313The ``xfbtree_trans_commit`` and ``xfbtree_trans_cancel`` functions implement 2314this functionality as follows: 2315 23161. Find each buffer log item whose buffer targets the xfile. 2317 23182. Record the dirty/ordered status of the log item. 2319 23203. Detach the log item from the buffer. 2321 23224. Queue the buffer to a special delwri list. 2323 23245. Clear the transaction dirty flag if the only dirty log items were the ones 2325 that were detached in step 3. 2326 23276. Submit the delwri list to commit the changes to the xfile, if the updates 2328 are being committed. 2329 2330After removing xfile logged buffers from the transaction in this manner, the 2331transaction can be committed or cancelled. 2332 2333Bulk Loading of Ondisk B+Trees 2334------------------------------ 2335 2336As mentioned previously, early iterations of online repair built new btree 2337structures by creating a new btree and adding observations individually. 2338Loading a btree one record at a time had a slight advantage of not requiring 2339the incore records to be sorted prior to commit, but was very slow and leaked 2340blocks if the system went down during a repair. 2341Loading records one at a time also meant that repair could not control the 2342loading factor of the blocks in the new btree. 2343 2344Fortunately, the venerable ``xfs_repair`` tool had a more efficient means for 2345rebuilding a btree index from a collection of records -- bulk btree loading. 2346This was implemented rather inefficiently code-wise, since ``xfs_repair`` 2347had separate copy-pasted implementations for each btree type. 2348 2349To prepare for online fsck, each of the four bulk loaders were studied, notes 2350were taken, and the four were refactored into a single generic btree bulk 2351loading mechanism. 2352Those notes in turn have been refreshed and are presented below. 2353 2354Geometry Computation 2355```````````````````` 2356 2357The zeroth step of bulk loading is to assemble the entire record set that will 2358be stored in the new btree, and sort the records. 2359Next, call ``xfs_btree_bload_compute_geometry`` to compute the shape of the 2360btree from the record set, the type of btree, and any load factor preferences. 2361This information is required for resource reservation. 2362 2363First, the geometry computation computes the minimum and maximum records that 2364will fit in a leaf block from the size of a btree block and the size of the 2365block header. 2366Roughly speaking, the maximum number of records is:: 2367 2368 maxrecs = (block_size - header_size) / record_size 2369 2370The XFS design specifies that btree blocks should be merged when possible, 2371which means the minimum number of records is half of maxrecs:: 2372 2373 minrecs = maxrecs / 2 2374 2375The next variable to determine is the desired loading factor. 2376This must be at least minrecs and no more than maxrecs. 2377Choosing minrecs is undesirable because it wastes half the block. 2378Choosing maxrecs is also undesirable because adding a single record to each 2379newly rebuilt leaf block will cause a tree split, which causes a noticeable 2380drop in performance immediately afterwards. 2381The default loading factor was chosen to be 75% of maxrecs, which provides a 2382reasonably compact structure without any immediate split penalties:: 2383 2384 default_load_factor = (maxrecs + minrecs) / 2 2385 2386If space is tight, the loading factor will be set to maxrecs to try to avoid 2387running out of space:: 2388 2389 leaf_load_factor = enough space ? default_load_factor : maxrecs 2390 2391Load factor is computed for btree node blocks using the combined size of the 2392btree key and pointer as the record size:: 2393 2394 maxrecs = (block_size - header_size) / (key_size + ptr_size) 2395 minrecs = maxrecs / 2 2396 node_load_factor = enough space ? default_load_factor : maxrecs 2397 2398Once that's done, the number of leaf blocks required to store the record set 2399can be computed as:: 2400 2401 leaf_blocks = ceil(record_count / leaf_load_factor) 2402 2403The number of node blocks needed to point to the next level down in the tree 2404is computed as:: 2405 2406 n_blocks = (n == 0 ? leaf_blocks : node_blocks[n]) 2407 node_blocks[n + 1] = ceil(n_blocks / node_load_factor) 2408 2409The entire computation is performed recursively until the current level only 2410needs one block. 2411The resulting geometry is as follows: 2412 2413- For AG-rooted btrees, this level is the root level, so the height of the new 2414 tree is ``level + 1`` and the space needed is the summation of the number of 2415 blocks on each level. 2416 2417- For inode-rooted btrees where the records in the top level do not fit in the 2418 inode fork area, the height is ``level + 2``, the space needed is the 2419 summation of the number of blocks on each level, and the inode fork points to 2420 the root block. 2421 2422- For inode-rooted btrees where the records in the top level can be stored in 2423 the inode fork area, then the root block can be stored in the inode, the 2424 height is ``level + 1``, and the space needed is one less than the summation 2425 of the number of blocks on each level. 2426 This only becomes relevant when non-bmap btrees gain the ability to root in 2427 an inode, which is a future patchset and only included here for completeness. 2428 2429.. _newbt: 2430 2431Reserving New B+Tree Blocks 2432``````````````````````````` 2433 2434Once repair knows the number of blocks needed for the new btree, it allocates 2435those blocks using the free space information. 2436Each reserved extent is tracked separately by the btree builder state data. 2437To improve crash resilience, the reservation code also logs an Extent Freeing 2438Intent (EFI) item in the same transaction as each space allocation and attaches 2439its in-memory ``struct xfs_extent_free_item`` object to the space reservation. 2440If the system goes down, log recovery will use the unfinished EFIs to free the 2441unused space, the free space, leaving the filesystem unchanged. 2442 2443Each time the btree builder claims a block for the btree from a reserved 2444extent, it updates the in-memory reservation to reflect the claimed space. 2445Block reservation tries to allocate as much contiguous space as possible to 2446reduce the number of EFIs in play. 2447 2448While repair is writing these new btree blocks, the EFIs created for the space 2449reservations pin the tail of the ondisk log. 2450It's possible that other parts of the system will remain busy and push the head 2451of the log towards the pinned tail. 2452To avoid livelocking the filesystem, the EFIs must not pin the tail of the log 2453for too long. 2454To alleviate this problem, the dynamic relogging capability of the deferred ops 2455mechanism is reused here to commit a transaction at the log head containing an 2456EFD for the old EFI and new EFI at the head. 2457This enables the log to release the old EFI to keep the log moving forwards. 2458 2459EFIs have a role to play during the commit and reaping phases; please see the 2460next section and the section about :ref:`reaping<reaping>` for more details. 2461 2462Proposed patchsets are the 2463`bitmap rework 2464<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-bitmap-rework>`_ 2465and the 2466`preparation for bulk loading btrees 2467<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-prep-for-bulk-loading>`_. 2468 2469 2470Writing the New Tree 2471```````````````````` 2472 2473This part is pretty simple -- the btree builder (``xfs_btree_bulkload``) claims 2474a block from the reserved list, writes the new btree block header, fills the 2475rest of the block with records, and adds the new leaf block to a list of 2476written blocks:: 2477 2478 ┌────┐ 2479 │leaf│ 2480 │RRR │ 2481 └────┘ 2482 2483Sibling pointers are set every time a new block is added to the level:: 2484 2485 ┌────┐ ┌────┐ ┌────┐ ┌────┐ 2486 │leaf│→│leaf│→│leaf│→│leaf│ 2487 │RRR │←│RRR │←│RRR │←│RRR │ 2488 └────┘ └────┘ └────┘ └────┘ 2489 2490When it finishes writing the record leaf blocks, it moves on to the node 2491blocks 2492To fill a node block, it walks each block in the next level down in the tree 2493to compute the relevant keys and write them into the parent node:: 2494 2495 ┌────┐ ┌────┐ 2496 │node│──────→│node│ 2497 │PP │←──────│PP │ 2498 └────┘ └────┘ 2499 ↙ ↘ ↙ ↘ 2500 ┌────┐ ┌────┐ ┌────┐ ┌────┐ 2501 │leaf│→│leaf│→│leaf│→│leaf│ 2502 │RRR │←│RRR │←│RRR │←│RRR │ 2503 └────┘ └────┘ └────┘ └────┘ 2504 2505When it reaches the root level, it is ready to commit the new btree!:: 2506 2507 ┌─────────┐ 2508 │ root │ 2509 │ PP │ 2510 └─────────┘ 2511 ↙ ↘ 2512 ┌────┐ ┌────┐ 2513 │node│──────→│node│ 2514 │PP │←──────│PP │ 2515 └────┘ └────┘ 2516 ↙ ↘ ↙ ↘ 2517 ┌────┐ ┌────┐ ┌────┐ ┌────┐ 2518 │leaf│→│leaf│→│leaf│→│leaf│ 2519 │RRR │←│RRR │←│RRR │←│RRR │ 2520 └────┘ └────┘ └────┘ └────┘ 2521 2522The first step to commit the new btree is to persist the btree blocks to disk 2523synchronously. 2524This is a little complicated because a new btree block could have been freed 2525in the recent past, so the builder must use ``xfs_buf_delwri_queue_here`` to 2526remove the (stale) buffer from the AIL list before it can write the new blocks 2527to disk. 2528Blocks are queued for IO using a delwri list and written in one large batch 2529with ``xfs_buf_delwri_submit``. 2530 2531Once the new blocks have been persisted to disk, control returns to the 2532individual repair function that called the bulk loader. 2533The repair function must log the location of the new root in a transaction, 2534clean up the space reservations that were made for the new btree, and reap the 2535old metadata blocks: 2536 25371. Commit the location of the new btree root. 2538 25392. For each incore reservation: 2540 2541 a. Log Extent Freeing Done (EFD) items for all the space that was consumed 2542 by the btree builder. The new EFDs must point to the EFIs attached to 2543 the reservation to prevent log recovery from freeing the new blocks. 2544 2545 b. For unclaimed portions of incore reservations, create a regular deferred 2546 extent free work item to be free the unused space later in the 2547 transaction chain. 2548 2549 c. The EFDs and EFIs logged in steps 2a and 2b must not overrun the 2550 reservation of the committing transaction. 2551 If the btree loading code suspects this might be about to happen, it must 2552 call ``xrep_defer_finish`` to clear out the deferred work and obtain a 2553 fresh transaction. 2554 25553. Clear out the deferred work a second time to finish the commit and clean 2556 the repair transaction. 2557 2558The transaction rolling in steps 2c and 3 represent a weakness in the repair 2559algorithm, because a log flush and a crash before the end of the reap step can 2560result in space leaking. 2561Online repair functions minimize the chances of this occurring by using very 2562large transactions, which each can accommodate many thousands of block freeing 2563instructions. 2564Repair moves on to reaping the old blocks, which will be presented in a 2565subsequent :ref:`section<reaping>` after a few case studies of bulk loading. 2566 2567Case Study: Rebuilding the Inode Index 2568^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2569 2570The high level process to rebuild the inode index btree is: 2571 25721. Walk the reverse mapping records to generate ``struct xfs_inobt_rec`` 2573 records from the inode chunk information and a bitmap of the old inode btree 2574 blocks. 2575 25762. Append the records to an xfarray in inode order. 2577 25783. Use the ``xfs_btree_bload_compute_geometry`` function to compute the number 2579 of blocks needed for the inode btree. 2580 If the free space inode btree is enabled, call it again to estimate the 2581 geometry of the finobt. 2582 25834. Allocate the number of blocks computed in the previous step. 2584 25855. Use ``xfs_btree_bload`` to write the xfarray records to btree blocks and 2586 generate the internal node blocks. 2587 If the free space inode btree is enabled, call it again to load the finobt. 2588 25896. Commit the location of the new btree root block(s) to the AGI. 2590 25917. Reap the old btree blocks using the bitmap created in step 1. 2592 2593Details are as follows. 2594 2595The inode btree maps inumbers to the ondisk location of the associated 2596inode records, which means that the inode btrees can be rebuilt from the 2597reverse mapping information. 2598Reverse mapping records with an owner of ``XFS_RMAP_OWN_INOBT`` marks the 2599location of the old inode btree blocks. 2600Each reverse mapping record with an owner of ``XFS_RMAP_OWN_INODES`` marks the 2601location of at least one inode cluster buffer. 2602A cluster is the smallest number of ondisk inodes that can be allocated or 2603freed in a single transaction; it is never smaller than 1 fs block or 4 inodes. 2604 2605For the space represented by each inode cluster, ensure that there are no 2606records in the free space btrees nor any records in the reference count btree. 2607If there are, the space metadata inconsistencies are reason enough to abort the 2608operation. 2609Otherwise, read each cluster buffer to check that its contents appear to be 2610ondisk inodes and to decide if the file is allocated 2611(``xfs_dinode.i_mode != 0``) or free (``xfs_dinode.i_mode == 0``). 2612Accumulate the results of successive inode cluster buffer reads until there is 2613enough information to fill a single inode chunk record, which is 64 consecutive 2614numbers in the inumber keyspace. 2615If the chunk is sparse, the chunk record may include holes. 2616 2617Once the repair function accumulates one chunk's worth of data, it calls 2618``xfarray_append`` to add the inode btree record to the xfarray. 2619This xfarray is walked twice during the btree creation step -- once to populate 2620the inode btree with all inode chunk records, and a second time to populate the 2621free inode btree with records for chunks that have free non-sparse inodes. 2622The number of records for the inode btree is the number of xfarray records, 2623but the record count for the free inode btree has to be computed as inode chunk 2624records are stored in the xfarray. 2625 2626The proposed patchset is the 2627`AG btree repair 2628<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-ag-btrees>`_ 2629series. 2630 2631Case Study: Rebuilding the Space Reference Counts 2632^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2633 2634Reverse mapping records are used to rebuild the reference count information. 2635Reference counts are required for correct operation of copy on write for shared 2636file data. 2637Imagine the reverse mapping entries as rectangles representing extents of 2638physical blocks, and that the rectangles can be laid down to allow them to 2639overlap each other. 2640From the diagram below, it is apparent that a reference count record must start 2641or end wherever the height of the stack changes. 2642In other words, the record emission stimulus is level-triggered:: 2643 2644 █ ███ 2645 ██ █████ ████ ███ ██████ 2646 ██ ████ ███████████ ████ █████████ 2647 ████████████████████████████████ ███████████ 2648 ^ ^ ^^ ^^ ^ ^^ ^^^ ^^^^ ^ ^^ ^ ^ ^ 2649 2 1 23 21 3 43 234 2123 1 01 2 3 0 2650 2651The ondisk reference count btree does not store the refcount == 0 cases because 2652the free space btree already records which blocks are free. 2653Extents being used to stage copy-on-write operations should be the only records 2654with refcount == 1. 2655Single-owner file blocks aren't recorded in either the free space or the 2656reference count btrees. 2657 2658The high level process to rebuild the reference count btree is: 2659 26601. Walk the reverse mapping records to generate ``struct xfs_refcount_irec`` 2661 records for any space having more than one reverse mapping and add them to 2662 the xfarray. 2663 Any records owned by ``XFS_RMAP_OWN_COW`` are also added to the xfarray 2664 because these are extents allocated to stage a copy on write operation and 2665 are tracked in the refcount btree. 2666 2667 Use any records owned by ``XFS_RMAP_OWN_REFC`` to create a bitmap of old 2668 refcount btree blocks. 2669 26702. Sort the records in physical extent order, putting the CoW staging extents 2671 at the end of the xfarray. 2672 This matches the sorting order of records in the refcount btree. 2673 26743. Use the ``xfs_btree_bload_compute_geometry`` function to compute the number 2675 of blocks needed for the new tree. 2676 26774. Allocate the number of blocks computed in the previous step. 2678 26795. Use ``xfs_btree_bload`` to write the xfarray records to btree blocks and 2680 generate the internal node blocks. 2681 26826. Commit the location of new btree root block to the AGF. 2683 26847. Reap the old btree blocks using the bitmap created in step 1. 2685 2686Details are as follows; the same algorithm is used by ``xfs_repair`` to 2687generate refcount information from reverse mapping records. 2688 2689- Until the reverse mapping btree runs out of records: 2690 2691 - Retrieve the next record from the btree and put it in a bag. 2692 2693 - Collect all records with the same starting block from the btree and put 2694 them in the bag. 2695 2696 - While the bag isn't empty: 2697 2698 - Among the mappings in the bag, compute the lowest block number where the 2699 reference count changes. 2700 This position will be either the starting block number of the next 2701 unprocessed reverse mapping or the next block after the shortest mapping 2702 in the bag. 2703 2704 - Remove all mappings from the bag that end at this position. 2705 2706 - Collect all reverse mappings that start at this position from the btree 2707 and put them in the bag. 2708 2709 - If the size of the bag changed and is greater than one, create a new 2710 refcount record associating the block number range that we just walked to 2711 the size of the bag. 2712 2713The bag-like structure in this case is a type 2 xfarray as discussed in the 2714:ref:`xfarray access patterns<xfarray_access_patterns>` section. 2715Reverse mappings are added to the bag using ``xfarray_store_anywhere`` and 2716removed via ``xfarray_unset``. 2717Bag members are examined through ``xfarray_iter`` loops. 2718 2719The proposed patchset is the 2720`AG btree repair 2721<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-ag-btrees>`_ 2722series. 2723 2724Case Study: Rebuilding File Fork Mapping Indices 2725^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 2726 2727The high level process to rebuild a data/attr fork mapping btree is: 2728 27291. Walk the reverse mapping records to generate ``struct xfs_bmbt_rec`` 2730 records from the reverse mapping records for that inode and fork. 2731 Append these records to an xfarray. 2732 Compute the bitmap of the old bmap btree blocks from the ``BMBT_BLOCK`` 2733 records. 2734 27352. Use the ``xfs_btree_bload_compute_geometry`` function to compute the number 2736 of blocks needed for the new tree. 2737 27383. Sort the records in file offset order. 2739 27404. If the extent records would fit in the inode fork immediate area, commit the 2741 records to that immediate area and skip to step 8. 2742 27435. Allocate the number of blocks computed in the previous step. 2744 27456. Use ``xfs_btree_bload`` to write the xfarray records to btree blocks and 2746 generate the internal node blocks. 2747 27487. Commit the new btree root block to the inode fork immediate area. 2749 27508. Reap the old btree blocks using the bitmap created in step 1. 2751 2752There are some complications here: 2753First, it's possible to move the fork offset to adjust the sizes of the 2754immediate areas if the data and attr forks are not both in BMBT format. 2755Second, if there are sufficiently few fork mappings, it may be possible to use 2756EXTENTS format instead of BMBT, which may require a conversion. 2757Third, the incore extent map must be reloaded carefully to avoid disturbing 2758any delayed allocation extents. 2759 2760The proposed patchset is the 2761`file mapping repair 2762<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-file-mappings>`_ 2763series. 2764 2765.. _reaping: 2766 2767Reaping Old Metadata Blocks 2768--------------------------- 2769 2770Whenever online fsck builds a new data structure to replace one that is 2771suspect, there is a question of how to find and dispose of the blocks that 2772belonged to the old structure. 2773The laziest method of course is not to deal with them at all, but this slowly 2774leads to service degradations as space leaks out of the filesystem. 2775Hopefully, someone will schedule a rebuild of the free space information to 2776plug all those leaks. 2777Offline repair rebuilds all space metadata after recording the usage of 2778the files and directories that it decides not to clear, hence it can build new 2779structures in the discovered free space and avoid the question of reaping. 2780 2781As part of a repair, online fsck relies heavily on the reverse mapping records 2782to find space that is owned by the corresponding rmap owner yet truly free. 2783Cross referencing rmap records with other rmap records is necessary because 2784there may be other data structures that also think they own some of those 2785blocks (e.g. crosslinked trees). 2786Permitting the block allocator to hand them out again will not push the system 2787towards consistency. 2788 2789For space metadata, the process of finding extents to dispose of generally 2790follows this format: 2791 27921. Create a bitmap of space used by data structures that must be preserved. 2793 The space reservations used to create the new metadata can be used here if 2794 the same rmap owner code is used to denote all of the objects being rebuilt. 2795 27962. Survey the reverse mapping data to create a bitmap of space owned by the 2797 same ``XFS_RMAP_OWN_*`` number for the metadata that is being preserved. 2798 27993. Use the bitmap disunion operator to subtract (1) from (2). 2800 The remaining set bits represent candidate extents that could be freed. 2801 The process moves on to step 4 below. 2802 2803Repairs for file-based metadata such as extended attributes, directories, 2804symbolic links, quota files and realtime bitmaps are performed by building a 2805new structure attached to a temporary file and swapping the forks. 2806Afterward, the mappings in the old file fork are the candidate blocks for 2807disposal. 2808 2809The process for disposing of old extents is as follows: 2810 28114. For each candidate extent, count the number of reverse mapping records for 2812 the first block in that extent that do not have the same rmap owner for the 2813 data structure being repaired. 2814 2815 - If zero, the block has a single owner and can be freed. 2816 2817 - If not, the block is part of a crosslinked structure and must not be 2818 freed. 2819 28205. Starting with the next block in the extent, figure out how many more blocks 2821 have the same zero/nonzero other owner status as that first block. 2822 28236. If the region is crosslinked, delete the reverse mapping entry for the 2824 structure being repaired and move on to the next region. 2825 28267. If the region is to be freed, mark any corresponding buffers in the buffer 2827 cache as stale to prevent log writeback. 2828 28298. Free the region and move on. 2830 2831However, there is one complication to this procedure. 2832Transactions are of finite size, so the reaping process must be careful to roll 2833the transactions to avoid overruns. 2834Overruns come from two sources: 2835 2836a. EFIs logged on behalf of space that is no longer occupied 2837 2838b. Log items for buffer invalidations 2839 2840This is also a window in which a crash during the reaping process can leak 2841blocks. 2842As stated earlier, online repair functions use very large transactions to 2843minimize the chances of this occurring. 2844 2845The proposed patchset is the 2846`preparation for bulk loading btrees 2847<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-prep-for-bulk-loading>`_ 2848series. 2849 2850Case Study: Reaping After a Regular Btree Repair 2851```````````````````````````````````````````````` 2852 2853Old reference count and inode btrees are the easiest to reap because they have 2854rmap records with special owner codes: ``XFS_RMAP_OWN_REFC`` for the refcount 2855btree, and ``XFS_RMAP_OWN_INOBT`` for the inode and free inode btrees. 2856Creating a list of extents to reap the old btree blocks is quite simple, 2857conceptually: 2858 28591. Lock the relevant AGI/AGF header buffers to prevent allocation and frees. 2860 28612. For each reverse mapping record with an rmap owner corresponding to the 2862 metadata structure being rebuilt, set the corresponding range in a bitmap. 2863 28643. Walk the current data structures that have the same rmap owner. 2865 For each block visited, clear that range in the above bitmap. 2866 28674. Each set bit in the bitmap represents a block that could be a block from the 2868 old data structures and hence is a candidate for reaping. 2869 In other words, ``(rmap_records_owned_by & ~blocks_reachable_by_walk)`` 2870 are the blocks that might be freeable. 2871 2872If it is possible to maintain the AGF lock throughout the repair (which is the 2873common case), then step 2 can be performed at the same time as the reverse 2874mapping record walk that creates the records for the new btree. 2875 2876Case Study: Rebuilding the Free Space Indices 2877````````````````````````````````````````````` 2878 2879The high level process to rebuild the free space indices is: 2880 28811. Walk the reverse mapping records to generate ``struct xfs_alloc_rec_incore`` 2882 records from the gaps in the reverse mapping btree. 2883 28842. Append the records to an xfarray. 2885 28863. Use the ``xfs_btree_bload_compute_geometry`` function to compute the number 2887 of blocks needed for each new tree. 2888 28894. Allocate the number of blocks computed in the previous step from the free 2890 space information collected. 2891 28925. Use ``xfs_btree_bload`` to write the xfarray records to btree blocks and 2893 generate the internal node blocks for the free space by length index. 2894 Call it again for the free space by block number index. 2895 28966. Commit the locations of the new btree root blocks to the AGF. 2897 28987. Reap the old btree blocks by looking for space that is not recorded by the 2899 reverse mapping btree, the new free space btrees, or the AGFL. 2900 2901Repairing the free space btrees has three key complications over a regular 2902btree repair: 2903 2904First, free space is not explicitly tracked in the reverse mapping records. 2905Hence, the new free space records must be inferred from gaps in the physical 2906space component of the keyspace of the reverse mapping btree. 2907 2908Second, free space repairs cannot use the common btree reservation code because 2909new blocks are reserved out of the free space btrees. 2910This is impossible when repairing the free space btrees themselves. 2911However, repair holds the AGF buffer lock for the duration of the free space 2912index reconstruction, so it can use the collected free space information to 2913supply the blocks for the new free space btrees. 2914It is not necessary to back each reserved extent with an EFI because the new 2915free space btrees are constructed in what the ondisk filesystem thinks is 2916unowned space. 2917However, if reserving blocks for the new btrees from the collected free space 2918information changes the number of free space records, repair must re-estimate 2919the new free space btree geometry with the new record count until the 2920reservation is sufficient. 2921As part of committing the new btrees, repair must ensure that reverse mappings 2922are created for the reserved blocks and that unused reserved blocks are 2923inserted into the free space btrees. 2924Deferrred rmap and freeing operations are used to ensure that this transition 2925is atomic, similar to the other btree repair functions. 2926 2927Third, finding the blocks to reap after the repair is not overly 2928straightforward. 2929Blocks for the free space btrees and the reverse mapping btrees are supplied by 2930the AGFL. 2931Blocks put onto the AGFL have reverse mapping records with the owner 2932``XFS_RMAP_OWN_AG``. 2933This ownership is retained when blocks move from the AGFL into the free space 2934btrees or the reverse mapping btrees. 2935When repair walks reverse mapping records to synthesize free space records, it 2936creates a bitmap (``ag_owner_bitmap``) of all the space claimed by 2937``XFS_RMAP_OWN_AG`` records. 2938The repair context maintains a second bitmap corresponding to the rmap btree 2939blocks and the AGFL blocks (``rmap_agfl_bitmap``). 2940When the walk is complete, the bitmap disunion operation ``(ag_owner_bitmap & 2941~rmap_agfl_bitmap)`` computes the extents that are used by the old free space 2942btrees. 2943These blocks can then be reaped using the methods outlined above. 2944 2945The proposed patchset is the 2946`AG btree repair 2947<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-ag-btrees>`_ 2948series. 2949 2950.. _rmap_reap: 2951 2952Case Study: Reaping After Repairing Reverse Mapping Btrees 2953`````````````````````````````````````````````````````````` 2954 2955Old reverse mapping btrees are less difficult to reap after a repair. 2956As mentioned in the previous section, blocks on the AGFL, the two free space 2957btree blocks, and the reverse mapping btree blocks all have reverse mapping 2958records with ``XFS_RMAP_OWN_AG`` as the owner. 2959The full process of gathering reverse mapping records and building a new btree 2960are described in the case study of 2961:ref:`live rebuilds of rmap data <rmap_repair>`, but a crucial point from that 2962discussion is that the new rmap btree will not contain any records for the old 2963rmap btree, nor will the old btree blocks be tracked in the free space btrees. 2964The list of candidate reaping blocks is computed by setting the bits 2965corresponding to the gaps in the new rmap btree records, and then clearing the 2966bits corresponding to extents in the free space btrees and the current AGFL 2967blocks. 2968The result ``(new_rmapbt_gaps & ~(agfl | bnobt_records))`` are reaped using the 2969methods outlined above. 2970 2971The rest of the process of rebuildng the reverse mapping btree is discussed 2972in a separate :ref:`case study<rmap_repair>`. 2973 2974The proposed patchset is the 2975`AG btree repair 2976<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-ag-btrees>`_ 2977series. 2978 2979Case Study: Rebuilding the AGFL 2980``````````````````````````````` 2981 2982The allocation group free block list (AGFL) is repaired as follows: 2983 29841. Create a bitmap for all the space that the reverse mapping data claims is 2985 owned by ``XFS_RMAP_OWN_AG``. 2986 29872. Subtract the space used by the two free space btrees and the rmap btree. 2988 29893. Subtract any space that the reverse mapping data claims is owned by any 2990 other owner, to avoid re-adding crosslinked blocks to the AGFL. 2991 29924. Once the AGFL is full, reap any blocks leftover. 2993 29945. The next operation to fix the freelist will right-size the list. 2995 2996See `fs/xfs/scrub/agheader_repair.c <https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/fs/xfs/scrub/agheader_repair.c>`_ for more details. 2997 2998Inode Record Repairs 2999-------------------- 3000 3001Inode records must be handled carefully, because they have both ondisk records 3002("dinodes") and an in-memory ("cached") representation. 3003There is a very high potential for cache coherency issues if online fsck is not 3004careful to access the ondisk metadata *only* when the ondisk metadata is so 3005badly damaged that the filesystem cannot load the in-memory representation. 3006When online fsck wants to open a damaged file for scrubbing, it must use 3007specialized resource acquisition functions that return either the in-memory 3008representation *or* a lock on whichever object is necessary to prevent any 3009update to the ondisk location. 3010 3011The only repairs that should be made to the ondisk inode buffers are whatever 3012is necessary to get the in-core structure loaded. 3013This means fixing whatever is caught by the inode cluster buffer and inode fork 3014verifiers, and retrying the ``iget`` operation. 3015If the second ``iget`` fails, the repair has failed. 3016 3017Once the in-memory representation is loaded, repair can lock the inode and can 3018subject it to comprehensive checks, repairs, and optimizations. 3019Most inode attributes are easy to check and constrain, or are user-controlled 3020arbitrary bit patterns; these are both easy to fix. 3021Dealing with the data and attr fork extent counts and the file block counts is 3022more complicated, because computing the correct value requires traversing the 3023forks, or if that fails, leaving the fields invalid and waiting for the fork 3024fsck functions to run. 3025 3026The proposed patchset is the 3027`inode 3028<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-inodes>`_ 3029repair series. 3030 3031Quota Record Repairs 3032-------------------- 3033 3034Similar to inodes, quota records ("dquots") also have both ondisk records and 3035an in-memory representation, and hence are subject to the same cache coherency 3036issues. 3037Somewhat confusingly, both are known as dquots in the XFS codebase. 3038 3039The only repairs that should be made to the ondisk quota record buffers are 3040whatever is necessary to get the in-core structure loaded. 3041Once the in-memory representation is loaded, the only attributes needing 3042checking are obviously bad limits and timer values. 3043 3044Quota usage counters are checked, repaired, and discussed separately in the 3045section about :ref:`live quotacheck <quotacheck>`. 3046 3047The proposed patchset is the 3048`quota 3049<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-quota>`_ 3050repair series. 3051 3052.. _fscounters: 3053 3054Freezing to Fix Summary Counters 3055-------------------------------- 3056 3057Filesystem summary counters track availability of filesystem resources such 3058as free blocks, free inodes, and allocated inodes. 3059This information could be compiled by walking the free space and inode indexes, 3060but this is a slow process, so XFS maintains a copy in the ondisk superblock 3061that should reflect the ondisk metadata, at least when the filesystem has been 3062unmounted cleanly. 3063For performance reasons, XFS also maintains incore copies of those counters, 3064which are key to enabling resource reservations for active transactions. 3065Writer threads reserve the worst-case quantities of resources from the 3066incore counter and give back whatever they don't use at commit time. 3067It is therefore only necessary to serialize on the superblock when the 3068superblock is being committed to disk. 3069 3070The lazy superblock counter feature introduced in XFS v5 took this even further 3071by training log recovery to recompute the summary counters from the AG headers, 3072which eliminated the need for most transactions even to touch the superblock. 3073The only time XFS commits the summary counters is at filesystem unmount. 3074To reduce contention even further, the incore counter is implemented as a 3075percpu counter, which means that each CPU is allocated a batch of blocks from a 3076global incore counter and can satisfy small allocations from the local batch. 3077 3078The high-performance nature of the summary counters makes it difficult for 3079online fsck to check them, since there is no way to quiesce a percpu counter 3080while the system is running. 3081Although online fsck can read the filesystem metadata to compute the correct 3082values of the summary counters, there's no way to hold the value of a percpu 3083counter stable, so it's quite possible that the counter will be out of date by 3084the time the walk is complete. 3085Earlier versions of online scrub would return to userspace with an incomplete 3086scan flag, but this is not a satisfying outcome for a system administrator. 3087For repairs, the in-memory counters must be stabilized while walking the 3088filesystem metadata to get an accurate reading and install it in the percpu 3089counter. 3090 3091To satisfy this requirement, online fsck must prevent other programs in the 3092system from initiating new writes to the filesystem, it must disable background 3093garbage collection threads, and it must wait for existing writer programs to 3094exit the kernel. 3095Once that has been established, scrub can walk the AG free space indexes, the 3096inode btrees, and the realtime bitmap to compute the correct value of all 3097four summary counters. 3098This is very similar to a filesystem freeze, though not all of the pieces are 3099necessary: 3100 3101- The final freeze state is set one higher than ``SB_FREEZE_COMPLETE`` to 3102 prevent other threads from thawing the filesystem, or other scrub threads 3103 from initiating another fscounters freeze. 3104 3105- It does not quiesce the log. 3106 3107With this code in place, it is now possible to pause the filesystem for just 3108long enough to check and correct the summary counters. 3109 3110+--------------------------------------------------------------------------+ 3111| **Historical Sidebar**: | 3112+--------------------------------------------------------------------------+ 3113| The initial implementation used the actual VFS filesystem freeze | 3114| mechanism to quiesce filesystem activity. | 3115| With the filesystem frozen, it is possible to resolve the counter values | 3116| with exact precision, but there are many problems with calling the VFS | 3117| methods directly: | 3118| | 3119| - Other programs can unfreeze the filesystem without our knowledge. | 3120| This leads to incorrect scan results and incorrect repairs. | 3121| | 3122| - Adding an extra lock to prevent others from thawing the filesystem | 3123| required the addition of a ``->freeze_super`` function to wrap | 3124| ``freeze_fs()``. | 3125| This in turn caused other subtle problems because it turns out that | 3126| the VFS ``freeze_super`` and ``thaw_super`` functions can drop the | 3127| last reference to the VFS superblock, and any subsequent access | 3128| becomes a UAF bug! | 3129| This can happen if the filesystem is unmounted while the underlying | 3130| block device has frozen the filesystem. | 3131| This problem could be solved by grabbing extra references to the | 3132| superblock, but it felt suboptimal given the other inadequacies of | 3133| this approach. | 3134| | 3135| - The log need not be quiesced to check the summary counters, but a VFS | 3136| freeze initiates one anyway. | 3137| This adds unnecessary runtime to live fscounter fsck operations. | 3138| | 3139| - Quiescing the log means that XFS flushes the (possibly incorrect) | 3140| counters to disk as part of cleaning the log. | 3141| | 3142| - A bug in the VFS meant that freeze could complete even when | 3143| sync_filesystem fails to flush the filesystem and returns an error. | 3144| This bug was fixed in Linux 5.17. | 3145+--------------------------------------------------------------------------+ 3146 3147The proposed patchset is the 3148`summary counter cleanup 3149<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-fscounters>`_ 3150series. 3151 3152Full Filesystem Scans 3153--------------------- 3154 3155Certain types of metadata can only be checked by walking every file in the 3156entire filesystem to record observations and comparing the observations against 3157what's recorded on disk. 3158Like every other type of online repair, repairs are made by writing those 3159observations to disk in a replacement structure and committing it atomically. 3160However, it is not practical to shut down the entire filesystem to examine 3161hundreds of billions of files because the downtime would be excessive. 3162Therefore, online fsck must build the infrastructure to manage a live scan of 3163all the files in the filesystem. 3164There are two questions that need to be solved to perform a live walk: 3165 3166- How does scrub manage the scan while it is collecting data? 3167 3168- How does the scan keep abreast of changes being made to the system by other 3169 threads? 3170 3171.. _iscan: 3172 3173Coordinated Inode Scans 3174``````````````````````` 3175 3176In the original Unix filesystems of the 1970s, each directory entry contained 3177an index number (*inumber*) which was used as an index into on ondisk array 3178(*itable*) of fixed-size records (*inodes*) describing a file's attributes and 3179its data block mapping. 3180This system is described by J. Lions, `"inode (5659)" 3181<http://www.lemis.com/grog/Documentation/Lions/>`_ in *Lions' Commentary on 3182UNIX, 6th Edition*, (Dept. of Computer Science, the University of New South 3183Wales, November 1977), pp. 18-2; and later by D. Ritchie and K. Thompson, 3184`"Implementation of the File System" 3185<https://archive.org/details/bstj57-6-1905/page/n8/mode/1up>`_, from *The UNIX 3186Time-Sharing System*, (The Bell System Technical Journal, July 1978), pp. 31871913-4. 3188 3189XFS retains most of this design, except now inumbers are search keys over all 3190the space in the data section filesystem. 3191They form a continuous keyspace that can be expressed as a 64-bit integer, 3192though the inodes themselves are sparsely distributed within the keyspace. 3193Scans proceed in a linear fashion across the inumber keyspace, starting from 3194``0x0`` and ending at ``0xFFFFFFFFFFFFFFFF``. 3195Naturally, a scan through a keyspace requires a scan cursor object to track the 3196scan progress. 3197Because this keyspace is sparse, this cursor contains two parts. 3198The first part of this scan cursor object tracks the inode that will be 3199examined next; call this the examination cursor. 3200Somewhat less obviously, the scan cursor object must also track which parts of 3201the keyspace have already been visited, which is critical for deciding if a 3202concurrent filesystem update needs to be incorporated into the scan data. 3203Call this the visited inode cursor. 3204 3205Advancing the scan cursor is a multi-step process encapsulated in 3206``xchk_iscan_iter``: 3207 32081. Lock the AGI buffer of the AG containing the inode pointed to by the visited 3209 inode cursor. 3210 This guarantee that inodes in this AG cannot be allocated or freed while 3211 advancing the cursor. 3212 32132. Use the per-AG inode btree to look up the next inumber after the one that 3214 was just visited, since it may not be keyspace adjacent. 3215 32163. If there are no more inodes left in this AG: 3217 3218 a. Move the examination cursor to the point of the inumber keyspace that 3219 corresponds to the start of the next AG. 3220 3221 b. Adjust the visited inode cursor to indicate that it has "visited" the 3222 last possible inode in the current AG's inode keyspace. 3223 XFS inumbers are segmented, so the cursor needs to be marked as having 3224 visited the entire keyspace up to just before the start of the next AG's 3225 inode keyspace. 3226 3227 c. Unlock the AGI and return to step 1 if there are unexamined AGs in the 3228 filesystem. 3229 3230 d. If there are no more AGs to examine, set both cursors to the end of the 3231 inumber keyspace. 3232 The scan is now complete. 3233 32344. Otherwise, there is at least one more inode to scan in this AG: 3235 3236 a. Move the examination cursor ahead to the next inode marked as allocated 3237 by the inode btree. 3238 3239 b. Adjust the visited inode cursor to point to the inode just prior to where 3240 the examination cursor is now. 3241 Because the scanner holds the AGI buffer lock, no inodes could have been 3242 created in the part of the inode keyspace that the visited inode cursor 3243 just advanced. 3244 32455. Get the incore inode for the inumber of the examination cursor. 3246 By maintaining the AGI buffer lock until this point, the scanner knows that 3247 it was safe to advance the examination cursor across the entire keyspace, 3248 and that it has stabilized this next inode so that it cannot disappear from 3249 the filesystem until the scan releases the incore inode. 3250 32516. Drop the AGI lock and return the incore inode to the caller. 3252 3253Online fsck functions scan all files in the filesystem as follows: 3254 32551. Start a scan by calling ``xchk_iscan_start``. 3256 32572. Advance the scan cursor (``xchk_iscan_iter``) to get the next inode. 3258 If one is provided: 3259 3260 a. Lock the inode to prevent updates during the scan. 3261 3262 b. Scan the inode. 3263 3264 c. While still holding the inode lock, adjust the visited inode cursor 3265 (``xchk_iscan_mark_visited``) to point to this inode. 3266 3267 d. Unlock and release the inode. 3268 32698. Call ``xchk_iscan_teardown`` to complete the scan. 3270 3271There are subtleties with the inode cache that complicate grabbing the incore 3272inode for the caller. 3273Obviously, it is an absolute requirement that the inode metadata be consistent 3274enough to load it into the inode cache. 3275Second, if the incore inode is stuck in some intermediate state, the scan 3276coordinator must release the AGI and push the main filesystem to get the inode 3277back into a loadable state. 3278 3279The proposed patches are the 3280`inode scanner 3281<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=scrub-iscan>`_ 3282series. 3283The first user of the new functionality is the 3284`online quotacheck 3285<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-quotacheck>`_ 3286series. 3287 3288Inode Management 3289```````````````` 3290 3291In regular filesystem code, references to allocated XFS incore inodes are 3292always obtained (``xfs_iget``) outside of transaction context because the 3293creation of the incore context for an existing file does not require metadata 3294updates. 3295However, it is important to note that references to incore inodes obtained as 3296part of file creation must be performed in transaction context because the 3297filesystem must ensure the atomicity of the ondisk inode btree index updates 3298and the initialization of the actual ondisk inode. 3299 3300References to incore inodes are always released (``xfs_irele``) outside of 3301transaction context because there are a handful of activities that might 3302require ondisk updates: 3303 3304- The VFS may decide to kick off writeback as part of a ``DONTCACHE`` inode 3305 release. 3306 3307- Speculative preallocations need to be unreserved. 3308 3309- An unlinked file may have lost its last reference, in which case the entire 3310 file must be inactivated, which involves releasing all of its resources in 3311 the ondisk metadata and freeing the inode. 3312 3313These activities are collectively called inode inactivation. 3314Inactivation has two parts -- the VFS part, which initiates writeback on all 3315dirty file pages, and the XFS part, which cleans up XFS-specific information 3316and frees the inode if it was unlinked. 3317If the inode is unlinked (or unconnected after a file handle operation), the 3318kernel drops the inode into the inactivation machinery immediately. 3319 3320During normal operation, resource acquisition for an update follows this order 3321to avoid deadlocks: 3322 33231. Inode reference (``iget``). 3324 33252. Filesystem freeze protection, if repairing (``mnt_want_write_file``). 3326 33273. Inode ``IOLOCK`` (VFS ``i_rwsem``) lock to control file IO. 3328 33294. Inode ``MMAPLOCK`` (page cache ``invalidate_lock``) lock for operations that 3330 can update page cache mappings. 3331 33325. Log feature enablement. 3333 33346. Transaction log space grant. 3335 33367. Space on the data and realtime devices for the transaction. 3337 33388. Incore dquot references, if a file is being repaired. 3339 Note that they are not locked, merely acquired. 3340 33419. Inode ``ILOCK`` for file metadata updates. 3342 334310. AG header buffer locks / Realtime metadata inode ILOCK. 3344 334511. Realtime metadata buffer locks, if applicable. 3346 334712. Extent mapping btree blocks, if applicable. 3348 3349Resources are often released in the reverse order, though this is not required. 3350However, online fsck differs from regular XFS operations because it may examine 3351an object that normally is acquired in a later stage of the locking order, and 3352then decide to cross-reference the object with an object that is acquired 3353earlier in the order. 3354The next few sections detail the specific ways in which online fsck takes care 3355to avoid deadlocks. 3356 3357iget and irele During a Scrub 3358^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 3359 3360An inode scan performed on behalf of a scrub operation runs in transaction 3361context, and possibly with resources already locked and bound to it. 3362This isn't much of a problem for ``iget`` since it can operate in the context 3363of an existing transaction, as long as all of the bound resources are acquired 3364before the inode reference in the regular filesystem. 3365 3366When the VFS ``iput`` function is given a linked inode with no other 3367references, it normally puts the inode on an LRU list in the hope that it can 3368save time if another process re-opens the file before the system runs out 3369of memory and frees it. 3370Filesystem callers can short-circuit the LRU process by setting a ``DONTCACHE`` 3371flag on the inode to cause the kernel to try to drop the inode into the 3372inactivation machinery immediately. 3373 3374In the past, inactivation was always done from the process that dropped the 3375inode, which was a problem for scrub because scrub may already hold a 3376transaction, and XFS does not support nesting transactions. 3377On the other hand, if there is no scrub transaction, it is desirable to drop 3378otherwise unused inodes immediately to avoid polluting caches. 3379To capture these nuances, the online fsck code has a separate ``xchk_irele`` 3380function to set or clear the ``DONTCACHE`` flag to get the required release 3381behavior. 3382 3383Proposed patchsets include fixing 3384`scrub iget usage 3385<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=scrub-iget-fixes>`_ and 3386`dir iget usage 3387<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=scrub-dir-iget-fixes>`_. 3388 3389.. _ilocking: 3390 3391Locking Inodes 3392^^^^^^^^^^^^^^ 3393 3394In regular filesystem code, the VFS and XFS will acquire multiple IOLOCK locks 3395in a well-known order: parent → child when updating the directory tree, and 3396in numerical order of the addresses of their ``struct inode`` object otherwise. 3397For regular files, the MMAPLOCK can be acquired after the IOLOCK to stop page 3398faults. 3399If two MMAPLOCKs must be acquired, they are acquired in numerical order of 3400the addresses of their ``struct address_space`` objects. 3401Due to the structure of existing filesystem code, IOLOCKs and MMAPLOCKs must be 3402acquired before transactions are allocated. 3403If two ILOCKs must be acquired, they are acquired in inumber order. 3404 3405Inode lock acquisition must be done carefully during a coordinated inode scan. 3406Online fsck cannot abide these conventions, because for a directory tree 3407scanner, the scrub process holds the IOLOCK of the file being scanned and it 3408needs to take the IOLOCK of the file at the other end of the directory link. 3409If the directory tree is corrupt because it contains a cycle, ``xfs_scrub`` 3410cannot use the regular inode locking functions and avoid becoming trapped in an 3411ABBA deadlock. 3412 3413Solving both of these problems is straightforward -- any time online fsck 3414needs to take a second lock of the same class, it uses trylock to avoid an ABBA 3415deadlock. 3416If the trylock fails, scrub drops all inode locks and use trylock loops to 3417(re)acquire all necessary resources. 3418Trylock loops enable scrub to check for pending fatal signals, which is how 3419scrub avoids deadlocking the filesystem or becoming an unresponsive process. 3420However, trylock loops means that online fsck must be prepared to measure the 3421resource being scrubbed before and after the lock cycle to detect changes and 3422react accordingly. 3423 3424.. _dirparent: 3425 3426Case Study: Finding a Directory Parent 3427^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 3428 3429Consider the directory parent pointer repair code as an example. 3430Online fsck must verify that the dotdot dirent of a directory points up to a 3431parent directory, and that the parent directory contains exactly one dirent 3432pointing down to the child directory. 3433Fully validating this relationship (and repairing it if possible) requires a 3434walk of every directory on the filesystem while holding the child locked, and 3435while updates to the directory tree are being made. 3436The coordinated inode scan provides a way to walk the filesystem without the 3437possibility of missing an inode. 3438The child directory is kept locked to prevent updates to the dotdot dirent, but 3439if the scanner fails to lock a parent, it can drop and relock both the child 3440and the prospective parent. 3441If the dotdot entry changes while the directory is unlocked, then a move or 3442rename operation must have changed the child's parentage, and the scan can 3443exit early. 3444 3445The proposed patchset is the 3446`directory repair 3447<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-dirs>`_ 3448series. 3449 3450.. _fshooks: 3451 3452Filesystem Hooks 3453````````````````` 3454 3455The second piece of support that online fsck functions need during a full 3456filesystem scan is the ability to stay informed about updates being made by 3457other threads in the filesystem, since comparisons against the past are useless 3458in a dynamic environment. 3459Two pieces of Linux kernel infrastructure enable online fsck to monitor regular 3460filesystem operations: filesystem hooks and :ref:`static keys<jump_labels>`. 3461 3462Filesystem hooks convey information about an ongoing filesystem operation to 3463a downstream consumer. 3464In this case, the downstream consumer is always an online fsck function. 3465Because multiple fsck functions can run in parallel, online fsck uses the Linux 3466notifier call chain facility to dispatch updates to any number of interested 3467fsck processes. 3468Call chains are a dynamic list, which means that they can be configured at 3469run time. 3470Because these hooks are private to the XFS module, the information passed along 3471contains exactly what the checking function needs to update its observations. 3472 3473The current implementation of XFS hooks uses SRCU notifier chains to reduce the 3474impact to highly threaded workloads. 3475Regular blocking notifier chains use a rwsem and seem to have a much lower 3476overhead for single-threaded applications. 3477However, it may turn out that the combination of blocking chains and static 3478keys are a more performant combination; more study is needed here. 3479 3480The following pieces are necessary to hook a certain point in the filesystem: 3481 3482- A ``struct xfs_hooks`` object must be embedded in a convenient place such as 3483 a well-known incore filesystem object. 3484 3485- Each hook must define an action code and a structure containing more context 3486 about the action. 3487 3488- Hook providers should provide appropriate wrapper functions and structs 3489 around the ``xfs_hooks`` and ``xfs_hook`` objects to take advantage of type 3490 checking to ensure correct usage. 3491 3492- A callsite in the regular filesystem code must be chosen to call 3493 ``xfs_hooks_call`` with the action code and data structure. 3494 This place should be adjacent to (and not earlier than) the place where 3495 the filesystem update is committed to the transaction. 3496 In general, when the filesystem calls a hook chain, it should be able to 3497 handle sleeping and should not be vulnerable to memory reclaim or locking 3498 recursion. 3499 However, the exact requirements are very dependent on the context of the hook 3500 caller and the callee. 3501 3502- The online fsck function should define a structure to hold scan data, a lock 3503 to coordinate access to the scan data, and a ``struct xfs_hook`` object. 3504 The scanner function and the regular filesystem code must acquire resources 3505 in the same order; see the next section for details. 3506 3507- The online fsck code must contain a C function to catch the hook action code 3508 and data structure. 3509 If the object being updated has already been visited by the scan, then the 3510 hook information must be applied to the scan data. 3511 3512- Prior to unlocking inodes to start the scan, online fsck must call 3513 ``xfs_hooks_setup`` to initialize the ``struct xfs_hook``, and 3514 ``xfs_hooks_add`` to enable the hook. 3515 3516- Online fsck must call ``xfs_hooks_del`` to disable the hook once the scan is 3517 complete. 3518 3519The number of hooks should be kept to a minimum to reduce complexity. 3520Static keys are used to reduce the overhead of filesystem hooks to nearly 3521zero when online fsck is not running. 3522 3523.. _liveupdate: 3524 3525Live Updates During a Scan 3526`````````````````````````` 3527 3528The code paths of the online fsck scanning code and the :ref:`hooked<fshooks>` 3529filesystem code look like this:: 3530 3531 other program 3532 ↓ 3533 inode lock ←────────────────────┐ 3534 ↓ │ 3535 AG header lock │ 3536 ↓ │ 3537 filesystem function │ 3538 ↓ │ 3539 notifier call chain │ same 3540 ↓ ├─── inode 3541 scrub hook function │ lock 3542 ↓ │ 3543 scan data mutex ←──┐ same │ 3544 ↓ ├─── scan │ 3545 update scan data │ lock │ 3546 ↑ │ │ 3547 scan data mutex ←──┘ │ 3548 ↑ │ 3549 inode lock ←────────────────────┘ 3550 ↑ 3551 scrub function 3552 ↑ 3553 inode scanner 3554 ↑ 3555 xfs_scrub 3556 3557These rules must be followed to ensure correct interactions between the 3558checking code and the code making an update to the filesystem: 3559 3560- Prior to invoking the notifier call chain, the filesystem function being 3561 hooked must acquire the same lock that the scrub scanning function acquires 3562 to scan the inode. 3563 3564- The scanning function and the scrub hook function must coordinate access to 3565 the scan data by acquiring a lock on the scan data. 3566 3567- Scrub hook function must not add the live update information to the scan 3568 observations unless the inode being updated has already been scanned. 3569 The scan coordinator has a helper predicate (``xchk_iscan_want_live_update``) 3570 for this. 3571 3572- Scrub hook functions must not change the caller's state, including the 3573 transaction that it is running. 3574 They must not acquire any resources that might conflict with the filesystem 3575 function being hooked. 3576 3577- The hook function can abort the inode scan to avoid breaking the other rules. 3578 3579The inode scan APIs are pretty simple: 3580 3581- ``xchk_iscan_start`` starts a scan 3582 3583- ``xchk_iscan_iter`` grabs a reference to the next inode in the scan or 3584 returns zero if there is nothing left to scan 3585 3586- ``xchk_iscan_want_live_update`` to decide if an inode has already been 3587 visited in the scan. 3588 This is critical for hook functions to decide if they need to update the 3589 in-memory scan information. 3590 3591- ``xchk_iscan_mark_visited`` to mark an inode as having been visited in the 3592 scan 3593 3594- ``xchk_iscan_teardown`` to finish the scan 3595 3596This functionality is also a part of the 3597`inode scanner 3598<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=scrub-iscan>`_ 3599series. 3600 3601.. _quotacheck: 3602 3603Case Study: Quota Counter Checking 3604^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 3605 3606It is useful to compare the mount time quotacheck code to the online repair 3607quotacheck code. 3608Mount time quotacheck does not have to contend with concurrent operations, so 3609it does the following: 3610 36111. Make sure the ondisk dquots are in good enough shape that all the incore 3612 dquots will actually load, and zero the resource usage counters in the 3613 ondisk buffer. 3614 36152. Walk every inode in the filesystem. 3616 Add each file's resource usage to the incore dquot. 3617 36183. Walk each incore dquot. 3619 If the incore dquot is not being flushed, add the ondisk buffer backing the 3620 incore dquot to a delayed write (delwri) list. 3621 36224. Write the buffer list to disk. 3623 3624Like most online fsck functions, online quotacheck can't write to regular 3625filesystem objects until the newly collected metadata reflect all filesystem 3626state. 3627Therefore, online quotacheck records file resource usage to a shadow dquot 3628index implemented with a sparse ``xfarray``, and only writes to the real dquots 3629once the scan is complete. 3630Handling transactional updates is tricky because quota resource usage updates 3631are handled in phases to minimize contention on dquots: 3632 36331. The inodes involved are joined and locked to a transaction. 3634 36352. For each dquot attached to the file: 3636 3637 a. The dquot is locked. 3638 3639 b. A quota reservation is added to the dquot's resource usage. 3640 The reservation is recorded in the transaction. 3641 3642 c. The dquot is unlocked. 3643 36443. Changes in actual quota usage are tracked in the transaction. 3645 36464. At transaction commit time, each dquot is examined again: 3647 3648 a. The dquot is locked again. 3649 3650 b. Quota usage changes are logged and unused reservation is given back to 3651 the dquot. 3652 3653 c. The dquot is unlocked. 3654 3655For online quotacheck, hooks are placed in steps 2 and 4. 3656The step 2 hook creates a shadow version of the transaction dquot context 3657(``dqtrx``) that operates in a similar manner to the regular code. 3658The step 4 hook commits the shadow ``dqtrx`` changes to the shadow dquots. 3659Notice that both hooks are called with the inode locked, which is how the 3660live update coordinates with the inode scanner. 3661 3662The quotacheck scan looks like this: 3663 36641. Set up a coordinated inode scan. 3665 36662. For each inode returned by the inode scan iterator: 3667 3668 a. Grab and lock the inode. 3669 3670 b. Determine that inode's resource usage (data blocks, inode counts, 3671 realtime blocks) and add that to the shadow dquots for the user, group, 3672 and project ids associated with the inode. 3673 3674 c. Unlock and release the inode. 3675 36763. For each dquot in the system: 3677 3678 a. Grab and lock the dquot. 3679 3680 b. Check the dquot against the shadow dquots created by the scan and updated 3681 by the live hooks. 3682 3683Live updates are key to being able to walk every quota record without 3684needing to hold any locks for a long duration. 3685If repairs are desired, the real and shadow dquots are locked and their 3686resource counts are set to the values in the shadow dquot. 3687 3688The proposed patchset is the 3689`online quotacheck 3690<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-quotacheck>`_ 3691series. 3692 3693.. _nlinks: 3694 3695Case Study: File Link Count Checking 3696^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 3697 3698File link count checking also uses live update hooks. 3699The coordinated inode scanner is used to visit all directories on the 3700filesystem, and per-file link count records are stored in a sparse ``xfarray`` 3701indexed by inumber. 3702During the scanning phase, each entry in a directory generates observation 3703data as follows: 3704 37051. If the entry is a dotdot (``'..'``) entry of the root directory, the 3706 directory's parent link count is bumped because the root directory's dotdot 3707 entry is self referential. 3708 37092. If the entry is a dotdot entry of a subdirectory, the parent's backref 3710 count is bumped. 3711 37123. If the entry is neither a dot nor a dotdot entry, the target file's parent 3713 count is bumped. 3714 37154. If the target is a subdirectory, the parent's child link count is bumped. 3716 3717A crucial point to understand about how the link count inode scanner interacts 3718with the live update hooks is that the scan cursor tracks which *parent* 3719directories have been scanned. 3720In other words, the live updates ignore any update about ``A → B`` when A has 3721not been scanned, even if B has been scanned. 3722Furthermore, a subdirectory A with a dotdot entry pointing back to B is 3723accounted as a backref counter in the shadow data for A, since child dotdot 3724entries affect the parent's link count. 3725Live update hooks are carefully placed in all parts of the filesystem that 3726create, change, or remove directory entries, since those operations involve 3727bumplink and droplink. 3728 3729For any file, the correct link count is the number of parents plus the number 3730of child subdirectories. 3731Non-directories never have children of any kind. 3732The backref information is used to detect inconsistencies in the number of 3733links pointing to child subdirectories and the number of dotdot entries 3734pointing back. 3735 3736After the scan completes, the link count of each file can be checked by locking 3737both the inode and the shadow data, and comparing the link counts. 3738A second coordinated inode scan cursor is used for comparisons. 3739Live updates are key to being able to walk every inode without needing to hold 3740any locks between inodes. 3741If repairs are desired, the inode's link count is set to the value in the 3742shadow information. 3743If no parents are found, the file must be :ref:`reparented <orphanage>` to the 3744orphanage to prevent the file from being lost forever. 3745 3746The proposed patchset is the 3747`file link count repair 3748<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=scrub-nlinks>`_ 3749series. 3750 3751.. _rmap_repair: 3752 3753Case Study: Rebuilding Reverse Mapping Records 3754^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 3755 3756Most repair functions follow the same pattern: lock filesystem resources, 3757walk the surviving ondisk metadata looking for replacement metadata records, 3758and use an :ref:`in-memory array <xfarray>` to store the gathered observations. 3759The primary advantage of this approach is the simplicity and modularity of the 3760repair code -- code and data are entirely contained within the scrub module, 3761do not require hooks in the main filesystem, and are usually the most efficient 3762in memory use. 3763A secondary advantage of this repair approach is atomicity -- once the kernel 3764decides a structure is corrupt, no other threads can access the metadata until 3765the kernel finishes repairing and revalidating the metadata. 3766 3767For repairs going on within a shard of the filesystem, these advantages 3768outweigh the delays inherent in locking the shard while repairing parts of the 3769shard. 3770Unfortunately, repairs to the reverse mapping btree cannot use the "standard" 3771btree repair strategy because it must scan every space mapping of every fork of 3772every file in the filesystem, and the filesystem cannot stop. 3773Therefore, rmap repair foregoes atomicity between scrub and repair. 3774It combines a :ref:`coordinated inode scanner <iscan>`, :ref:`live update hooks 3775<liveupdate>`, and an :ref:`in-memory rmap btree <xfbtree>` to complete the 3776scan for reverse mapping records. 3777 37781. Set up an xfbtree to stage rmap records. 3779 37802. While holding the locks on the AGI and AGF buffers acquired during the 3781 scrub, generate reverse mappings for all AG metadata: inodes, btrees, CoW 3782 staging extents, and the internal log. 3783 37843. Set up an inode scanner. 3785 37864. Hook into rmap updates for the AG being repaired so that the live scan data 3787 can receive updates to the rmap btree from the rest of the filesystem during 3788 the file scan. 3789 37905. For each space mapping found in either fork of each file scanned, 3791 decide if the mapping matches the AG of interest. 3792 If so: 3793 3794 a. Create a btree cursor for the in-memory btree. 3795 3796 b. Use the rmap code to add the record to the in-memory btree. 3797 3798 c. Use the :ref:`special commit function <xfbtree_commit>` to write the 3799 xfbtree changes to the xfile. 3800 38016. For each live update received via the hook, decide if the owner has already 3802 been scanned. 3803 If so, apply the live update into the scan data: 3804 3805 a. Create a btree cursor for the in-memory btree. 3806 3807 b. Replay the operation into the in-memory btree. 3808 3809 c. Use the :ref:`special commit function <xfbtree_commit>` to write the 3810 xfbtree changes to the xfile. 3811 This is performed with an empty transaction to avoid changing the 3812 caller's state. 3813 38147. When the inode scan finishes, create a new scrub transaction and relock the 3815 two AG headers. 3816 38178. Compute the new btree geometry using the number of rmap records in the 3818 shadow btree, like all other btree rebuilding functions. 3819 38209. Allocate the number of blocks computed in the previous step. 3821 382210. Perform the usual btree bulk loading and commit to install the new rmap 3823 btree. 3824 382511. Reap the old rmap btree blocks as discussed in the case study about how 3826 to :ref:`reap after rmap btree repair <rmap_reap>`. 3827 382812. Free the xfbtree now that it not needed. 3829 3830The proposed patchset is the 3831`rmap repair 3832<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-rmap-btree>`_ 3833series. 3834 3835Staging Repairs with Temporary Files on Disk 3836-------------------------------------------- 3837 3838XFS stores a substantial amount of metadata in file forks: directories, 3839extended attributes, symbolic link targets, free space bitmaps and summary 3840information for the realtime volume, and quota records. 3841File forks map 64-bit logical file fork space extents to physical storage space 3842extents, similar to how a memory management unit maps 64-bit virtual addresses 3843to physical memory addresses. 3844Therefore, file-based tree structures (such as directories and extended 3845attributes) use blocks mapped in the file fork offset address space that point 3846to other blocks mapped within that same address space, and file-based linear 3847structures (such as bitmaps and quota records) compute array element offsets in 3848the file fork offset address space. 3849 3850Because file forks can consume as much space as the entire filesystem, repairs 3851cannot be staged in memory, even when a paging scheme is available. 3852Therefore, online repair of file-based metadata createas a temporary file in 3853the XFS filesystem, writes a new structure at the correct offsets into the 3854temporary file, and atomically swaps the fork mappings (and hence the fork 3855contents) to commit the repair. 3856Once the repair is complete, the old fork can be reaped as necessary; if the 3857system goes down during the reap, the iunlink code will delete the blocks 3858during log recovery. 3859 3860**Note**: All space usage and inode indices in the filesystem *must* be 3861consistent to use a temporary file safely! 3862This dependency is the reason why online repair can only use pageable kernel 3863memory to stage ondisk space usage information. 3864 3865Swapping metadata extents with a temporary file requires the owner field of the 3866block headers to match the file being repaired and not the temporary file. The 3867directory, extended attribute, and symbolic link functions were all modified to 3868allow callers to specify owner numbers explicitly. 3869 3870There is a downside to the reaping process -- if the system crashes during the 3871reap phase and the fork extents are crosslinked, the iunlink processing will 3872fail because freeing space will find the extra reverse mappings and abort. 3873 3874Temporary files created for repair are similar to ``O_TMPFILE`` files created 3875by userspace. 3876They are not linked into a directory and the entire file will be reaped when 3877the last reference to the file is lost. 3878The key differences are that these files must have no access permission outside 3879the kernel at all, they must be specially marked to prevent them from being 3880opened by handle, and they must never be linked into the directory tree. 3881 3882+--------------------------------------------------------------------------+ 3883| **Historical Sidebar**: | 3884+--------------------------------------------------------------------------+ 3885| In the initial iteration of file metadata repair, the damaged metadata | 3886| blocks would be scanned for salvageable data; the extents in the file | 3887| fork would be reaped; and then a new structure would be built in its | 3888| place. | 3889| This strategy did not survive the introduction of the atomic repair | 3890| requirement expressed earlier in this document. | 3891| | 3892| The second iteration explored building a second structure at a high | 3893| offset in the fork from the salvage data, reaping the old extents, and | 3894| using a ``COLLAPSE_RANGE`` operation to slide the new extents into | 3895| place. | 3896| | 3897| This had many drawbacks: | 3898| | 3899| - Array structures are linearly addressed, and the regular filesystem | 3900| codebase does not have the concept of a linear offset that could be | 3901| applied to the record offset computation to build an alternate copy. | 3902| | 3903| - Extended attributes are allowed to use the entire attr fork offset | 3904| address space. | 3905| | 3906| - Even if repair could build an alternate copy of a data structure in a | 3907| different part of the fork address space, the atomic repair commit | 3908| requirement means that online repair would have to be able to perform | 3909| a log assisted ``COLLAPSE_RANGE`` operation to ensure that the old | 3910| structure was completely replaced. | 3911| | 3912| - A crash after construction of the secondary tree but before the range | 3913| collapse would leave unreachable blocks in the file fork. | 3914| This would likely confuse things further. | 3915| | 3916| - Reaping blocks after a repair is not a simple operation, and | 3917| initiating a reap operation from a restarted range collapse operation | 3918| during log recovery is daunting. | 3919| | 3920| - Directory entry blocks and quota records record the file fork offset | 3921| in the header area of each block. | 3922| An atomic range collapse operation would have to rewrite this part of | 3923| each block header. | 3924| Rewriting a single field in block headers is not a huge problem, but | 3925| it's something to be aware of. | 3926| | 3927| - Each block in a directory or extended attributes btree index contains | 3928| sibling and child block pointers. | 3929| Were the atomic commit to use a range collapse operation, each block | 3930| would have to be rewritten very carefully to preserve the graph | 3931| structure. | 3932| Doing this as part of a range collapse means rewriting a large number | 3933| of blocks repeatedly, which is not conducive to quick repairs. | 3934| | 3935| This lead to the introduction of temporary file staging. | 3936+--------------------------------------------------------------------------+ 3937 3938Using a Temporary File 3939`````````````````````` 3940 3941Online repair code should use the ``xrep_tempfile_create`` function to create a 3942temporary file inside the filesystem. 3943This allocates an inode, marks the in-core inode private, and attaches it to 3944the scrub context. 3945These files are hidden from userspace, may not be added to the directory tree, 3946and must be kept private. 3947 3948Temporary files only use two inode locks: the IOLOCK and the ILOCK. 3949The MMAPLOCK is not needed here, because there must not be page faults from 3950userspace for data fork blocks. 3951The usage patterns of these two locks are the same as for any other XFS file -- 3952access to file data are controlled via the IOLOCK, and access to file metadata 3953are controlled via the ILOCK. 3954Locking helpers are provided so that the temporary file and its lock state can 3955be cleaned up by the scrub context. 3956To comply with the nested locking strategy laid out in the :ref:`inode 3957locking<ilocking>` section, it is recommended that scrub functions use the 3958xrep_tempfile_ilock*_nowait lock helpers. 3959 3960Data can be written to a temporary file by two means: 3961 39621. ``xrep_tempfile_copyin`` can be used to set the contents of a regular 3963 temporary file from an xfile. 3964 39652. The regular directory, symbolic link, and extended attribute functions can 3966 be used to write to the temporary file. 3967 3968Once a good copy of a data file has been constructed in a temporary file, it 3969must be conveyed to the file being repaired, which is the topic of the next 3970section. 3971 3972The proposed patches are in the 3973`repair temporary files 3974<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-tempfiles>`_ 3975series. 3976 3977Atomic Extent Swapping 3978---------------------- 3979 3980Once repair builds a temporary file with a new data structure written into 3981it, it must commit the new changes into the existing file. 3982It is not possible to swap the inumbers of two files, so instead the new 3983metadata must replace the old. 3984This suggests the need for the ability to swap extents, but the existing extent 3985swapping code used by the file defragmenting tool ``xfs_fsr`` is not sufficient 3986for online repair because: 3987 3988a. When the reverse-mapping btree is enabled, the swap code must keep the 3989 reverse mapping information up to date with every exchange of mappings. 3990 Therefore, it can only exchange one mapping per transaction, and each 3991 transaction is independent. 3992 3993b. Reverse-mapping is critical for the operation of online fsck, so the old 3994 defragmentation code (which swapped entire extent forks in a single 3995 operation) is not useful here. 3996 3997c. Defragmentation is assumed to occur between two files with identical 3998 contents. 3999 For this use case, an incomplete exchange will not result in a user-visible 4000 change in file contents, even if the operation is interrupted. 4001 4002d. Online repair needs to swap the contents of two files that are by definition 4003 *not* identical. 4004 For directory and xattr repairs, the user-visible contents might be the 4005 same, but the contents of individual blocks may be very different. 4006 4007e. Old blocks in the file may be cross-linked with another structure and must 4008 not reappear if the system goes down mid-repair. 4009 4010These problems are overcome by creating a new deferred operation and a new type 4011of log intent item to track the progress of an operation to exchange two file 4012ranges. 4013The new deferred operation type chains together the same transactions used by 4014the reverse-mapping extent swap code. 4015The new log item records the progress of the exchange to ensure that once an 4016exchange begins, it will always run to completion, even there are 4017interruptions. 4018The new ``XFS_SB_FEAT_INCOMPAT_LOG_ATOMIC_SWAP`` log-incompatible feature flag 4019in the superblock protects these new log item records from being replayed on 4020old kernels. 4021 4022The proposed patchset is the 4023`atomic extent swap 4024<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=atomic-file-updates>`_ 4025series. 4026 4027+--------------------------------------------------------------------------+ 4028| **Sidebar: Using Log-Incompatible Feature Flags** | 4029+--------------------------------------------------------------------------+ 4030| Starting with XFS v5, the superblock contains a | 4031| ``sb_features_log_incompat`` field to indicate that the log contains | 4032| records that might not readable by all kernels that could mount this | 4033| filesystem. | 4034| In short, log incompat features protect the log contents against kernels | 4035| that will not understand the contents. | 4036| Unlike the other superblock feature bits, log incompat bits are | 4037| ephemeral because an empty (clean) log does not need protection. | 4038| The log cleans itself after its contents have been committed into the | 4039| filesystem, either as part of an unmount or because the system is | 4040| otherwise idle. | 4041| Because upper level code can be working on a transaction at the same | 4042| time that the log cleans itself, it is necessary for upper level code to | 4043| communicate to the log when it is going to use a log incompatible | 4044| feature. | 4045| | 4046| The log coordinates access to incompatible features through the use of | 4047| one ``struct rw_semaphore`` for each feature. | 4048| The log cleaning code tries to take this rwsem in exclusive mode to | 4049| clear the bit; if the lock attempt fails, the feature bit remains set. | 4050| Filesystem code signals its intention to use a log incompat feature in a | 4051| transaction by calling ``xlog_use_incompat_feat``, which takes the rwsem | 4052| in shared mode. | 4053| The code supporting a log incompat feature should create wrapper | 4054| functions to obtain the log feature and call | 4055| ``xfs_add_incompat_log_feature`` to set the feature bits in the primary | 4056| superblock. | 4057| The superblock update is performed transactionally, so the wrapper to | 4058| obtain log assistance must be called just prior to the creation of the | 4059| transaction that uses the functionality. | 4060| For a file operation, this step must happen after taking the IOLOCK | 4061| and the MMAPLOCK, but before allocating the transaction. | 4062| When the transaction is complete, the ``xlog_drop_incompat_feat`` | 4063| function is called to release the feature. | 4064| The feature bit will not be cleared from the superblock until the log | 4065| becomes clean. | 4066| | 4067| Log-assisted extended attribute updates and atomic extent swaps both use | 4068| log incompat features and provide convenience wrappers around the | 4069| functionality. | 4070+--------------------------------------------------------------------------+ 4071 4072Mechanics of an Atomic Extent Swap 4073`````````````````````````````````` 4074 4075Swapping entire file forks is a complex task. 4076The goal is to exchange all file fork mappings between two file fork offset 4077ranges. 4078There are likely to be many extent mappings in each fork, and the edges of 4079the mappings aren't necessarily aligned. 4080Furthermore, there may be other updates that need to happen after the swap, 4081such as exchanging file sizes, inode flags, or conversion of fork data to local 4082format. 4083This is roughly the format of the new deferred extent swap work item: 4084 4085.. code-block:: c 4086 4087 struct xfs_swapext_intent { 4088 /* Inodes participating in the operation. */ 4089 struct xfs_inode *sxi_ip1; 4090 struct xfs_inode *sxi_ip2; 4091 4092 /* File offset range information. */ 4093 xfs_fileoff_t sxi_startoff1; 4094 xfs_fileoff_t sxi_startoff2; 4095 xfs_filblks_t sxi_blockcount; 4096 4097 /* Set these file sizes after the operation, unless negative. */ 4098 xfs_fsize_t sxi_isize1; 4099 xfs_fsize_t sxi_isize2; 4100 4101 /* XFS_SWAP_EXT_* log operation flags */ 4102 uint64_t sxi_flags; 4103 }; 4104 4105The new log intent item contains enough information to track two logical fork 4106offset ranges: ``(inode1, startoff1, blockcount)`` and ``(inode2, startoff2, 4107blockcount)``. 4108Each step of a swap operation exchanges the largest file range mapping possible 4109from one file to the other. 4110After each step in the swap operation, the two startoff fields are incremented 4111and the blockcount field is decremented to reflect the progress made. 4112The flags field captures behavioral parameters such as swapping the attr fork 4113instead of the data fork and other work to be done after the extent swap. 4114The two isize fields are used to swap the file size at the end of the operation 4115if the file data fork is the target of the swap operation. 4116 4117When the extent swap is initiated, the sequence of operations is as follows: 4118 41191. Create a deferred work item for the extent swap. 4120 At the start, it should contain the entirety of the file ranges to be 4121 swapped. 4122 41232. Call ``xfs_defer_finish`` to process the exchange. 4124 This is encapsulated in ``xrep_tempswap_contents`` for scrub operations. 4125 This will log an extent swap intent item to the transaction for the deferred 4126 extent swap work item. 4127 41283. Until ``sxi_blockcount`` of the deferred extent swap work item is zero, 4129 4130 a. Read the block maps of both file ranges starting at ``sxi_startoff1`` and 4131 ``sxi_startoff2``, respectively, and compute the longest extent that can 4132 be swapped in a single step. 4133 This is the minimum of the two ``br_blockcount`` s in the mappings. 4134 Keep advancing through the file forks until at least one of the mappings 4135 contains written blocks. 4136 Mutual holes, unwritten extents, and extent mappings to the same physical 4137 space are not exchanged. 4138 4139 For the next few steps, this document will refer to the mapping that came 4140 from file 1 as "map1", and the mapping that came from file 2 as "map2". 4141 4142 b. Create a deferred block mapping update to unmap map1 from file 1. 4143 4144 c. Create a deferred block mapping update to unmap map2 from file 2. 4145 4146 d. Create a deferred block mapping update to map map1 into file 2. 4147 4148 e. Create a deferred block mapping update to map map2 into file 1. 4149 4150 f. Log the block, quota, and extent count updates for both files. 4151 4152 g. Extend the ondisk size of either file if necessary. 4153 4154 h. Log an extent swap done log item for the extent swap intent log item 4155 that was read at the start of step 3. 4156 4157 i. Compute the amount of file range that has just been covered. 4158 This quantity is ``(map1.br_startoff + map1.br_blockcount - 4159 sxi_startoff1)``, because step 3a could have skipped holes. 4160 4161 j. Increase the starting offsets of ``sxi_startoff1`` and ``sxi_startoff2`` 4162 by the number of blocks computed in the previous step, and decrease 4163 ``sxi_blockcount`` by the same quantity. 4164 This advances the cursor. 4165 4166 k. Log a new extent swap intent log item reflecting the advanced state of 4167 the work item. 4168 4169 l. Return the proper error code (EAGAIN) to the deferred operation manager 4170 to inform it that there is more work to be done. 4171 The operation manager completes the deferred work in steps 3b-3e before 4172 moving back to the start of step 3. 4173 41744. Perform any post-processing. 4175 This will be discussed in more detail in subsequent sections. 4176 4177If the filesystem goes down in the middle of an operation, log recovery will 4178find the most recent unfinished extent swap log intent item and restart from 4179there. 4180This is how extent swapping guarantees that an outside observer will either see 4181the old broken structure or the new one, and never a mismash of both. 4182 4183Preparation for Extent Swapping 4184``````````````````````````````` 4185 4186There are a few things that need to be taken care of before initiating an 4187atomic extent swap operation. 4188First, regular files require the page cache to be flushed to disk before the 4189operation begins, and directio writes to be quiesced. 4190Like any filesystem operation, extent swapping must determine the maximum 4191amount of disk space and quota that can be consumed on behalf of both files in 4192the operation, and reserve that quantity of resources to avoid an unrecoverable 4193out of space failure once it starts dirtying metadata. 4194The preparation step scans the ranges of both files to estimate: 4195 4196- Data device blocks needed to handle the repeated updates to the fork 4197 mappings. 4198- Change in data and realtime block counts for both files. 4199- Increase in quota usage for both files, if the two files do not share the 4200 same set of quota ids. 4201- The number of extent mappings that will be added to each file. 4202- Whether or not there are partially written realtime extents. 4203 User programs must never be able to access a realtime file extent that maps 4204 to different extents on the realtime volume, which could happen if the 4205 operation fails to run to completion. 4206 4207The need for precise estimation increases the run time of the swap operation, 4208but it is very important to maintain correct accounting. 4209The filesystem must not run completely out of free space, nor can the extent 4210swap ever add more extent mappings to a fork than it can support. 4211Regular users are required to abide the quota limits, though metadata repairs 4212may exceed quota to resolve inconsistent metadata elsewhere. 4213 4214Special Features for Swapping Metadata File Extents 4215``````````````````````````````````````````````````` 4216 4217Extended attributes, symbolic links, and directories can set the fork format to 4218"local" and treat the fork as a literal area for data storage. 4219Metadata repairs must take extra steps to support these cases: 4220 4221- If both forks are in local format and the fork areas are large enough, the 4222 swap is performed by copying the incore fork contents, logging both forks, 4223 and committing. 4224 The atomic extent swap mechanism is not necessary, since this can be done 4225 with a single transaction. 4226 4227- If both forks map blocks, then the regular atomic extent swap is used. 4228 4229- Otherwise, only one fork is in local format. 4230 The contents of the local format fork are converted to a block to perform the 4231 swap. 4232 The conversion to block format must be done in the same transaction that 4233 logs the initial extent swap intent log item. 4234 The regular atomic extent swap is used to exchange the mappings. 4235 Special flags are set on the swap operation so that the transaction can be 4236 rolled one more time to convert the second file's fork back to local format 4237 so that the second file will be ready to go as soon as the ILOCK is dropped. 4238 4239Extended attributes and directories stamp the owning inode into every block, 4240but the buffer verifiers do not actually check the inode number! 4241Although there is no verification, it is still important to maintain 4242referential integrity, so prior to performing the extent swap, online repair 4243builds every block in the new data structure with the owner field of the file 4244being repaired. 4245 4246After a successful swap operation, the repair operation must reap the old fork 4247blocks by processing each fork mapping through the standard :ref:`file extent 4248reaping <reaping>` mechanism that is done post-repair. 4249If the filesystem should go down during the reap part of the repair, the 4250iunlink processing at the end of recovery will free both the temporary file and 4251whatever blocks were not reaped. 4252However, this iunlink processing omits the cross-link detection of online 4253repair, and is not completely foolproof. 4254 4255Swapping Temporary File Extents 4256``````````````````````````````` 4257 4258To repair a metadata file, online repair proceeds as follows: 4259 42601. Create a temporary repair file. 4261 42622. Use the staging data to write out new contents into the temporary repair 4263 file. 4264 The same fork must be written to as is being repaired. 4265 42663. Commit the scrub transaction, since the swap estimation step must be 4267 completed before transaction reservations are made. 4268 42694. Call ``xrep_tempswap_trans_alloc`` to allocate a new scrub transaction with 4270 the appropriate resource reservations, locks, and fill out a ``struct 4271 xfs_swapext_req`` with the details of the swap operation. 4272 42735. Call ``xrep_tempswap_contents`` to swap the contents. 4274 42756. Commit the transaction to complete the repair. 4276 4277.. _rtsummary: 4278 4279Case Study: Repairing the Realtime Summary File 4280^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 4281 4282In the "realtime" section of an XFS filesystem, free space is tracked via a 4283bitmap, similar to Unix FFS. 4284Each bit in the bitmap represents one realtime extent, which is a multiple of 4285the filesystem block size between 4KiB and 1GiB in size. 4286The realtime summary file indexes the number of free extents of a given size to 4287the offset of the block within the realtime free space bitmap where those free 4288extents begin. 4289In other words, the summary file helps the allocator find free extents by 4290length, similar to what the free space by count (cntbt) btree does for the data 4291section. 4292 4293The summary file itself is a flat file (with no block headers or checksums!) 4294partitioned into ``log2(total rt extents)`` sections containing enough 32-bit 4295counters to match the number of blocks in the rt bitmap. 4296Each counter records the number of free extents that start in that bitmap block 4297and can satisfy a power-of-two allocation request. 4298 4299To check the summary file against the bitmap: 4300 43011. Take the ILOCK of both the realtime bitmap and summary files. 4302 43032. For each free space extent recorded in the bitmap: 4304 4305 a. Compute the position in the summary file that contains a counter that 4306 represents this free extent. 4307 4308 b. Read the counter from the xfile. 4309 4310 c. Increment it, and write it back to the xfile. 4311 43123. Compare the contents of the xfile against the ondisk file. 4313 4314To repair the summary file, write the xfile contents into the temporary file 4315and use atomic extent swap to commit the new contents. 4316The temporary file is then reaped. 4317 4318The proposed patchset is the 4319`realtime summary repair 4320<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-rtsummary>`_ 4321series. 4322 4323Case Study: Salvaging Extended Attributes 4324^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 4325 4326In XFS, extended attributes are implemented as a namespaced name-value store. 4327Values are limited in size to 64KiB, but there is no limit in the number of 4328names. 4329The attribute fork is unpartitioned, which means that the root of the attribute 4330structure is always in logical block zero, but attribute leaf blocks, dabtree 4331index blocks, and remote value blocks are intermixed. 4332Attribute leaf blocks contain variable-sized records that associate 4333user-provided names with the user-provided values. 4334Values larger than a block are allocated separate extents and written there. 4335If the leaf information expands beyond a single block, a directory/attribute 4336btree (``dabtree``) is created to map hashes of attribute names to entries 4337for fast lookup. 4338 4339Salvaging extended attributes is done as follows: 4340 43411. Walk the attr fork mappings of the file being repaired to find the attribute 4342 leaf blocks. 4343 When one is found, 4344 4345 a. Walk the attr leaf block to find candidate keys. 4346 When one is found, 4347 4348 1. Check the name for problems, and ignore the name if there are. 4349 4350 2. Retrieve the value. 4351 If that succeeds, add the name and value to the staging xfarray and 4352 xfblob. 4353 43542. If the memory usage of the xfarray and xfblob exceed a certain amount of 4355 memory or there are no more attr fork blocks to examine, unlock the file and 4356 add the staged extended attributes to the temporary file. 4357 43583. Use atomic extent swapping to exchange the new and old extended attribute 4359 structures. 4360 The old attribute blocks are now attached to the temporary file. 4361 43624. Reap the temporary file. 4363 4364The proposed patchset is the 4365`extended attribute repair 4366<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-xattrs>`_ 4367series. 4368 4369Fixing Directories 4370------------------ 4371 4372Fixing directories is difficult with currently available filesystem features, 4373since directory entries are not redundant. 4374The offline repair tool scans all inodes to find files with nonzero link count, 4375and then it scans all directories to establish parentage of those linked files. 4376Damaged files and directories are zapped, and files with no parent are 4377moved to the ``/lost+found`` directory. 4378It does not try to salvage anything. 4379 4380The best that online repair can do at this time is to read directory data 4381blocks and salvage any dirents that look plausible, correct link counts, and 4382move orphans back into the directory tree. 4383The salvage process is discussed in the case study at the end of this section. 4384The :ref:`file link count fsck <nlinks>` code takes care of fixing link counts 4385and moving orphans to the ``/lost+found`` directory. 4386 4387Case Study: Salvaging Directories 4388````````````````````````````````` 4389 4390Unlike extended attributes, directory blocks are all the same size, so 4391salvaging directories is straightforward: 4392 43931. Find the parent of the directory. 4394 If the dotdot entry is not unreadable, try to confirm that the alleged 4395 parent has a child entry pointing back to the directory being repaired. 4396 Otherwise, walk the filesystem to find it. 4397 43982. Walk the first partition of data fork of the directory to find the directory 4399 entry data blocks. 4400 When one is found, 4401 4402 a. Walk the directory data block to find candidate entries. 4403 When an entry is found: 4404 4405 i. Check the name for problems, and ignore the name if there are. 4406 4407 ii. Retrieve the inumber and grab the inode. 4408 If that succeeds, add the name, inode number, and file type to the 4409 staging xfarray and xblob. 4410 44113. If the memory usage of the xfarray and xfblob exceed a certain amount of 4412 memory or there are no more directory data blocks to examine, unlock the 4413 directory and add the staged dirents into the temporary directory. 4414 Truncate the staging files. 4415 44164. Use atomic extent swapping to exchange the new and old directory structures. 4417 The old directory blocks are now attached to the temporary file. 4418 44195. Reap the temporary file. 4420 4421**Future Work Question**: Should repair revalidate the dentry cache when 4422rebuilding a directory? 4423 4424*Answer*: Yes, it should. 4425 4426In theory it is necessary to scan all dentry cache entries for a directory to 4427ensure that one of the following apply: 4428 44291. The cached dentry reflects an ondisk dirent in the new directory. 4430 44312. The cached dentry no longer has a corresponding ondisk dirent in the new 4432 directory and the dentry can be purged from the cache. 4433 44343. The cached dentry no longer has an ondisk dirent but the dentry cannot be 4435 purged. 4436 This is the problem case. 4437 4438Unfortunately, the current dentry cache design doesn't provide a means to walk 4439every child dentry of a specific directory, which makes this a hard problem. 4440There is no known solution. 4441 4442The proposed patchset is the 4443`directory repair 4444<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-dirs>`_ 4445series. 4446 4447Parent Pointers 4448``````````````` 4449 4450A parent pointer is a piece of file metadata that enables a user to locate the 4451file's parent directory without having to traverse the directory tree from the 4452root. 4453Without them, reconstruction of directory trees is hindered in much the same 4454way that the historic lack of reverse space mapping information once hindered 4455reconstruction of filesystem space metadata. 4456The parent pointer feature, however, makes total directory reconstruction 4457possible. 4458 4459XFS parent pointers include the dirent name and location of the entry within 4460the parent directory. 4461In other words, child files use extended attributes to store pointers to 4462parents in the form ``(parent_inum, parent_gen, dirent_pos) → (dirent_name)``. 4463The directory checking process can be strengthened to ensure that the target of 4464each dirent also contains a parent pointer pointing back to the dirent. 4465Likewise, each parent pointer can be checked by ensuring that the target of 4466each parent pointer is a directory and that it contains a dirent matching 4467the parent pointer. 4468Both online and offline repair can use this strategy. 4469 4470**Note**: The ondisk format of parent pointers is not yet finalized. 4471 4472+--------------------------------------------------------------------------+ 4473| **Historical Sidebar**: | 4474+--------------------------------------------------------------------------+ 4475| Directory parent pointers were first proposed as an XFS feature more | 4476| than a decade ago by SGI. | 4477| Each link from a parent directory to a child file is mirrored with an | 4478| extended attribute in the child that could be used to identify the | 4479| parent directory. | 4480| Unfortunately, this early implementation had major shortcomings and was | 4481| never merged into Linux XFS: | 4482| | 4483| 1. The XFS codebase of the late 2000s did not have the infrastructure to | 4484| enforce strong referential integrity in the directory tree. | 4485| It did not guarantee that a change in a forward link would always be | 4486| followed up with the corresponding change to the reverse links. | 4487| | 4488| 2. Referential integrity was not integrated into offline repair. | 4489| Checking and repairs were performed on mounted filesystems without | 4490| taking any kernel or inode locks to coordinate access. | 4491| It is not clear how this actually worked properly. | 4492| | 4493| 3. The extended attribute did not record the name of the directory entry | 4494| in the parent, so the SGI parent pointer implementation cannot be | 4495| used to reconnect the directory tree. | 4496| | 4497| 4. Extended attribute forks only support 65,536 extents, which means | 4498| that parent pointer attribute creation is likely to fail at some | 4499| point before the maximum file link count is achieved. | 4500| | 4501| The original parent pointer design was too unstable for something like | 4502| a file system repair to depend on. | 4503| Allison Henderson, Chandan Babu, and Catherine Hoang are working on a | 4504| second implementation that solves all shortcomings of the first. | 4505| During 2022, Allison introduced log intent items to track physical | 4506| manipulations of the extended attribute structures. | 4507| This solves the referential integrity problem by making it possible to | 4508| commit a dirent update and a parent pointer update in the same | 4509| transaction. | 4510| Chandan increased the maximum extent counts of both data and attribute | 4511| forks, thereby ensuring that the extended attribute structure can grow | 4512| to handle the maximum hardlink count of any file. | 4513+--------------------------------------------------------------------------+ 4514 4515Case Study: Repairing Directories with Parent Pointers 4516^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 4517 4518Directory rebuilding uses a :ref:`coordinated inode scan <iscan>` and 4519a :ref:`directory entry live update hook <liveupdate>` as follows: 4520 45211. Set up a temporary directory for generating the new directory structure, 4522 an xfblob for storing entry names, and an xfarray for stashing directory 4523 updates. 4524 45252. Set up an inode scanner and hook into the directory entry code to receive 4526 updates on directory operations. 4527 45283. For each parent pointer found in each file scanned, decide if the parent 4529 pointer references the directory of interest. 4530 If so: 4531 4532 a. Stash an addname entry for this dirent in the xfarray for later. 4533 4534 b. When finished scanning that file, flush the stashed updates to the 4535 temporary directory. 4536 45374. For each live directory update received via the hook, decide if the child 4538 has already been scanned. 4539 If so: 4540 4541 a. Stash an addname or removename entry for this dirent update in the 4542 xfarray for later. 4543 We cannot write directly to the temporary directory because hook 4544 functions are not allowed to modify filesystem metadata. 4545 Instead, we stash updates in the xfarray and rely on the scanner thread 4546 to apply the stashed updates to the temporary directory. 4547 45485. When the scan is complete, atomically swap the contents of the temporary 4549 directory and the directory being repaired. 4550 The temporary directory now contains the damaged directory structure. 4551 45526. Reap the temporary directory. 4553 45547. Update the dirent position field of parent pointers as necessary. 4555 This may require the queuing of a substantial number of xattr log intent 4556 items. 4557 4558The proposed patchset is the 4559`parent pointers directory repair 4560<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=pptrs-online-dir-repair>`_ 4561series. 4562 4563**Unresolved Question**: How will repair ensure that the ``dirent_pos`` fields 4564match in the reconstructed directory? 4565 4566*Answer*: There are a few ways to solve this problem: 4567 45681. The field could be designated advisory, since the other three values are 4569 sufficient to find the entry in the parent. 4570 However, this makes indexed key lookup impossible while repairs are ongoing. 4571 45722. We could allow creating directory entries at specified offsets, which solves 4573 the referential integrity problem but runs the risk that dirent creation 4574 will fail due to conflicts with the free space in the directory. 4575 4576 These conflicts could be resolved by appending the directory entry and 4577 amending the xattr code to support updating an xattr key and reindexing the 4578 dabtree, though this would have to be performed with the parent directory 4579 still locked. 4580 45813. Same as above, but remove the old parent pointer entry and add a new one 4582 atomically. 4583 45844. Change the ondisk xattr format to ``(parent_inum, name) → (parent_gen)``, 4585 which would provide the attr name uniqueness that we require, without 4586 forcing repair code to update the dirent position. 4587 Unfortunately, this requires changes to the xattr code to support attr 4588 names as long as 263 bytes. 4589 45905. Change the ondisk xattr format to ``(parent_inum, hash(name)) → 4591 (name, parent_gen)``. 4592 If the hash is sufficiently resistant to collisions (e.g. sha256) then 4593 this should provide the attr name uniqueness that we require. 4594 Names shorter than 247 bytes could be stored directly. 4595 4596Discussion is ongoing under the `parent pointers patch deluge 4597<https://www.spinics.net/lists/linux-xfs/msg69397.html>`_. 4598 4599Case Study: Repairing Parent Pointers 4600^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 4601 4602Online reconstruction of a file's parent pointer information works similarly to 4603directory reconstruction: 4604 46051. Set up a temporary file for generating a new extended attribute structure, 4606 an `xfblob<xfblob>` for storing parent pointer names, and an xfarray for 4607 stashing parent pointer updates. 4608 46092. Set up an inode scanner and hook into the directory entry code to receive 4610 updates on directory operations. 4611 46123. For each directory entry found in each directory scanned, decide if the 4613 dirent references the file of interest. 4614 If so: 4615 4616 a. Stash an addpptr entry for this parent pointer in the xfblob and xfarray 4617 for later. 4618 4619 b. When finished scanning the directory, flush the stashed updates to the 4620 temporary directory. 4621 46224. For each live directory update received via the hook, decide if the parent 4623 has already been scanned. 4624 If so: 4625 4626 a. Stash an addpptr or removepptr entry for this dirent update in the 4627 xfarray for later. 4628 We cannot write parent pointers directly to the temporary file because 4629 hook functions are not allowed to modify filesystem metadata. 4630 Instead, we stash updates in the xfarray and rely on the scanner thread 4631 to apply the stashed parent pointer updates to the temporary file. 4632 46335. Copy all non-parent pointer extended attributes to the temporary file. 4634 46356. When the scan is complete, atomically swap the attribute fork of the 4636 temporary file and the file being repaired. 4637 The temporary file now contains the damaged extended attribute structure. 4638 46397. Reap the temporary file. 4640 4641The proposed patchset is the 4642`parent pointers repair 4643<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=pptrs-online-parent-repair>`_ 4644series. 4645 4646Digression: Offline Checking of Parent Pointers 4647^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 4648 4649Examining parent pointers in offline repair works differently because corrupt 4650files are erased long before directory tree connectivity checks are performed. 4651Parent pointer checks are therefore a second pass to be added to the existing 4652connectivity checks: 4653 46541. After the set of surviving files has been established (i.e. phase 6), 4655 walk the surviving directories of each AG in the filesystem. 4656 This is already performed as part of the connectivity checks. 4657 46582. For each directory entry found, record the name in an xfblob, and store 4659 ``(child_ag_inum, parent_inum, parent_gen, dirent_pos)`` tuples in a 4660 per-AG in-memory slab. 4661 46623. For each AG in the filesystem, 4663 4664 a. Sort the per-AG tuples in order of child_ag_inum, parent_inum, and 4665 dirent_pos. 4666 4667 b. For each inode in the AG, 4668 4669 1. Scan the inode for parent pointers. 4670 Record the names in a per-file xfblob, and store ``(parent_inum, 4671 parent_gen, dirent_pos)`` tuples in a per-file slab. 4672 4673 2. Sort the per-file tuples in order of parent_inum, and dirent_pos. 4674 4675 3. Position one slab cursor at the start of the inode's records in the 4676 per-AG tuple slab. 4677 This should be trivial since the per-AG tuples are in child inumber 4678 order. 4679 4680 4. Position a second slab cursor at the start of the per-file tuple slab. 4681 4682 5. Iterate the two cursors in lockstep, comparing the parent_ino and 4683 dirent_pos fields of the records under each cursor. 4684 4685 a. Tuples in the per-AG list but not the per-file list are missing and 4686 need to be written to the inode. 4687 4688 b. Tuples in the per-file list but not the per-AG list are dangling 4689 and need to be removed from the inode. 4690 4691 c. For tuples in both lists, update the parent_gen and name components 4692 of the parent pointer if necessary. 4693 46944. Move on to examining link counts, as we do today. 4695 4696The proposed patchset is the 4697`offline parent pointers repair 4698<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=pptrs-repair>`_ 4699series. 4700 4701Rebuilding directories from parent pointers in offline repair is very 4702challenging because it currently uses a single-pass scan of the filesystem 4703during phase 3 to decide which files are corrupt enough to be zapped. 4704This scan would have to be converted into a multi-pass scan: 4705 47061. The first pass of the scan zaps corrupt inodes, forks, and attributes 4707 much as it does now. 4708 Corrupt directories are noted but not zapped. 4709 47102. The next pass records parent pointers pointing to the directories noted 4711 as being corrupt in the first pass. 4712 This second pass may have to happen after the phase 4 scan for duplicate 4713 blocks, if phase 4 is also capable of zapping directories. 4714 47153. The third pass resets corrupt directories to an empty shortform directory. 4716 Free space metadata has not been ensured yet, so repair cannot yet use the 4717 directory building code in libxfs. 4718 47194. At the start of phase 6, space metadata have been rebuilt. 4720 Use the parent pointer information recorded during step 2 to reconstruct 4721 the dirents and add them to the now-empty directories. 4722 4723This code has not yet been constructed. 4724 4725.. _orphanage: 4726 4727The Orphanage 4728------------- 4729 4730Filesystems present files as a directed, and hopefully acyclic, graph. 4731In other words, a tree. 4732The root of the filesystem is a directory, and each entry in a directory points 4733downwards either to more subdirectories or to non-directory files. 4734Unfortunately, a disruption in the directory graph pointers result in a 4735disconnected graph, which makes files impossible to access via regular path 4736resolution. 4737 4738Without parent pointers, the directory parent pointer online scrub code can 4739detect a dotdot entry pointing to a parent directory that doesn't have a link 4740back to the child directory and the file link count checker can detect a file 4741that isn't pointed to by any directory in the filesystem. 4742If such a file has a positive link count, the file is an orphan. 4743 4744With parent pointers, directories can be rebuilt by scanning parent pointers 4745and parent pointers can be rebuilt by scanning directories. 4746This should reduce the incidence of files ending up in ``/lost+found``. 4747 4748When orphans are found, they should be reconnected to the directory tree. 4749Offline fsck solves the problem by creating a directory ``/lost+found`` to 4750serve as an orphanage, and linking orphan files into the orphanage by using the 4751inumber as the name. 4752Reparenting a file to the orphanage does not reset any of its permissions or 4753ACLs. 4754 4755This process is more involved in the kernel than it is in userspace. 4756The directory and file link count repair setup functions must use the regular 4757VFS mechanisms to create the orphanage directory with all the necessary 4758security attributes and dentry cache entries, just like a regular directory 4759tree modification. 4760 4761Orphaned files are adopted by the orphanage as follows: 4762 47631. Call ``xrep_orphanage_try_create`` at the start of the scrub setup function 4764 to try to ensure that the lost and found directory actually exists. 4765 This also attaches the orphanage directory to the scrub context. 4766 47672. If the decision is made to reconnect a file, take the IOLOCK of both the 4768 orphanage and the file being reattached. 4769 The ``xrep_orphanage_iolock_two`` function follows the inode locking 4770 strategy discussed earlier. 4771 47723. Call ``xrep_orphanage_compute_blkres`` and ``xrep_orphanage_compute_name`` 4773 to compute the new name in the orphanage and the block reservation required. 4774 47754. Use ``xrep_orphanage_adoption_prep`` to reserve resources to the repair 4776 transaction. 4777 47785. Call ``xrep_orphanage_adopt`` to reparent the orphaned file into the lost 4779 and found, and update the kernel dentry cache. 4780 4781The proposed patches are in the 4782`orphanage adoption 4783<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=repair-orphanage>`_ 4784series. 4785 47866. Userspace Algorithms and Data Structures 4787=========================================== 4788 4789This section discusses the key algorithms and data structures of the userspace 4790program, ``xfs_scrub``, that provide the ability to drive metadata checks and 4791repairs in the kernel, verify file data, and look for other potential problems. 4792 4793.. _scrubcheck: 4794 4795Checking Metadata 4796----------------- 4797 4798Recall the :ref:`phases of fsck work<scrubphases>` outlined earlier. 4799That structure follows naturally from the data dependencies designed into the 4800filesystem from its beginnings in 1993. 4801In XFS, there are several groups of metadata dependencies: 4802 4803a. Filesystem summary counts depend on consistency within the inode indices, 4804 the allocation group space btrees, and the realtime volume space 4805 information. 4806 4807b. Quota resource counts depend on consistency within the quota file data 4808 forks, inode indices, inode records, and the forks of every file on the 4809 system. 4810 4811c. The naming hierarchy depends on consistency within the directory and 4812 extended attribute structures. 4813 This includes file link counts. 4814 4815d. Directories, extended attributes, and file data depend on consistency within 4816 the file forks that map directory and extended attribute data to physical 4817 storage media. 4818 4819e. The file forks depends on consistency within inode records and the space 4820 metadata indices of the allocation groups and the realtime volume. 4821 This includes quota and realtime metadata files. 4822 4823f. Inode records depends on consistency within the inode metadata indices. 4824 4825g. Realtime space metadata depend on the inode records and data forks of the 4826 realtime metadata inodes. 4827 4828h. The allocation group metadata indices (free space, inodes, reference count, 4829 and reverse mapping btrees) depend on consistency within the AG headers and 4830 between all the AG metadata btrees. 4831 4832i. ``xfs_scrub`` depends on the filesystem being mounted and kernel support 4833 for online fsck functionality. 4834 4835Therefore, a metadata dependency graph is a convenient way to schedule checking 4836operations in the ``xfs_scrub`` program: 4837 4838- Phase 1 checks that the provided path maps to an XFS filesystem and detect 4839 the kernel's scrubbing abilities, which validates group (i). 4840 4841- Phase 2 scrubs groups (g) and (h) in parallel using a threaded workqueue. 4842 4843- Phase 3 scans inodes in parallel. 4844 For each inode, groups (f), (e), and (d) are checked, in that order. 4845 4846- Phase 4 repairs everything in groups (i) through (d) so that phases 5 and 6 4847 may run reliably. 4848 4849- Phase 5 starts by checking groups (b) and (c) in parallel before moving on 4850 to checking names. 4851 4852- Phase 6 depends on groups (i) through (b) to find file data blocks to verify, 4853 to read them, and to report which blocks of which files are affected. 4854 4855- Phase 7 checks group (a), having validated everything else. 4856 4857Notice that the data dependencies between groups are enforced by the structure 4858of the program flow. 4859 4860Parallel Inode Scans 4861-------------------- 4862 4863An XFS filesystem can easily contain hundreds of millions of inodes. 4864Given that XFS targets installations with large high-performance storage, 4865it is desirable to scrub inodes in parallel to minimize runtime, particularly 4866if the program has been invoked manually from a command line. 4867This requires careful scheduling to keep the threads as evenly loaded as 4868possible. 4869 4870Early iterations of the ``xfs_scrub`` inode scanner naïvely created a single 4871workqueue and scheduled a single workqueue item per AG. 4872Each workqueue item walked the inode btree (with ``XFS_IOC_INUMBERS``) to find 4873inode chunks and then called bulkstat (``XFS_IOC_BULKSTAT``) to gather enough 4874information to construct file handles. 4875The file handle was then passed to a function to generate scrub items for each 4876metadata object of each inode. 4877This simple algorithm leads to thread balancing problems in phase 3 if the 4878filesystem contains one AG with a few large sparse files and the rest of the 4879AGs contain many smaller files. 4880The inode scan dispatch function was not sufficiently granular; it should have 4881been dispatching at the level of individual inodes, or, to constrain memory 4882consumption, inode btree records. 4883 4884Thanks to Dave Chinner, bounded workqueues in userspace enable ``xfs_scrub`` to 4885avoid this problem with ease by adding a second workqueue. 4886Just like before, the first workqueue is seeded with one workqueue item per AG, 4887and it uses INUMBERS to find inode btree chunks. 4888The second workqueue, however, is configured with an upper bound on the number 4889of items that can be waiting to be run. 4890Each inode btree chunk found by the first workqueue's workers are queued to the 4891second workqueue, and it is this second workqueue that queries BULKSTAT, 4892creates a file handle, and passes it to a function to generate scrub items for 4893each metadata object of each inode. 4894If the second workqueue is too full, the workqueue add function blocks the 4895first workqueue's workers until the backlog eases. 4896This doesn't completely solve the balancing problem, but reduces it enough to 4897move on to more pressing issues. 4898 4899The proposed patchsets are the scrub 4900`performance tweaks 4901<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=scrub-performance-tweaks>`_ 4902and the 4903`inode scan rebalance 4904<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=scrub-iscan-rebalance>`_ 4905series. 4906 4907.. _scrubrepair: 4908 4909Scheduling Repairs 4910------------------ 4911 4912During phase 2, corruptions and inconsistencies reported in any AGI header or 4913inode btree are repaired immediately, because phase 3 relies on proper 4914functioning of the inode indices to find inodes to scan. 4915Failed repairs are rescheduled to phase 4. 4916Problems reported in any other space metadata are deferred to phase 4. 4917Optimization opportunities are always deferred to phase 4, no matter their 4918origin. 4919 4920During phase 3, corruptions and inconsistencies reported in any part of a 4921file's metadata are repaired immediately if all space metadata were validated 4922during phase 2. 4923Repairs that fail or cannot be repaired immediately are scheduled for phase 4. 4924 4925In the original design of ``xfs_scrub``, it was thought that repairs would be 4926so infrequent that the ``struct xfs_scrub_metadata`` objects used to 4927communicate with the kernel could also be used as the primary object to 4928schedule repairs. 4929With recent increases in the number of optimizations possible for a given 4930filesystem object, it became much more memory-efficient to track all eligible 4931repairs for a given filesystem object with a single repair item. 4932Each repair item represents a single lockable object -- AGs, metadata files, 4933individual inodes, or a class of summary information. 4934 4935Phase 4 is responsible for scheduling a lot of repair work in as quick a 4936manner as is practical. 4937The :ref:`data dependencies <scrubcheck>` outlined earlier still apply, which 4938means that ``xfs_scrub`` must try to complete the repair work scheduled by 4939phase 2 before trying repair work scheduled by phase 3. 4940The repair process is as follows: 4941 49421. Start a round of repair with a workqueue and enough workers to keep the CPUs 4943 as busy as the user desires. 4944 4945 a. For each repair item queued by phase 2, 4946 4947 i. Ask the kernel to repair everything listed in the repair item for a 4948 given filesystem object. 4949 4950 ii. Make a note if the kernel made any progress in reducing the number 4951 of repairs needed for this object. 4952 4953 iii. If the object no longer requires repairs, revalidate all metadata 4954 associated with this object. 4955 If the revalidation succeeds, drop the repair item. 4956 If not, requeue the item for more repairs. 4957 4958 b. If any repairs were made, jump back to 1a to retry all the phase 2 items. 4959 4960 c. For each repair item queued by phase 3, 4961 4962 i. Ask the kernel to repair everything listed in the repair item for a 4963 given filesystem object. 4964 4965 ii. Make a note if the kernel made any progress in reducing the number 4966 of repairs needed for this object. 4967 4968 iii. If the object no longer requires repairs, revalidate all metadata 4969 associated with this object. 4970 If the revalidation succeeds, drop the repair item. 4971 If not, requeue the item for more repairs. 4972 4973 d. If any repairs were made, jump back to 1c to retry all the phase 3 items. 4974 49752. If step 1 made any repair progress of any kind, jump back to step 1 to start 4976 another round of repair. 4977 49783. If there are items left to repair, run them all serially one more time. 4979 Complain if the repairs were not successful, since this is the last chance 4980 to repair anything. 4981 4982Corruptions and inconsistencies encountered during phases 5 and 7 are repaired 4983immediately. 4984Corrupt file data blocks reported by phase 6 cannot be recovered by the 4985filesystem. 4986 4987The proposed patchsets are the 4988`repair warning improvements 4989<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=scrub-better-repair-warnings>`_, 4990refactoring of the 4991`repair data dependency 4992<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=scrub-repair-data-deps>`_ 4993and 4994`object tracking 4995<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=scrub-object-tracking>`_, 4996and the 4997`repair scheduling 4998<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=scrub-repair-scheduling>`_ 4999improvement series. 5000 5001Checking Names for Confusable Unicode Sequences 5002----------------------------------------------- 5003 5004If ``xfs_scrub`` succeeds in validating the filesystem metadata by the end of 5005phase 4, it moves on to phase 5, which checks for suspicious looking names in 5006the filesystem. 5007These names consist of the filesystem label, names in directory entries, and 5008the names of extended attributes. 5009Like most Unix filesystems, XFS imposes the sparest of constraints on the 5010contents of a name: 5011 5012- Slashes and null bytes are not allowed in directory entries. 5013 5014- Null bytes are not allowed in userspace-visible extended attributes. 5015 5016- Null bytes are not allowed in the filesystem label. 5017 5018Directory entries and attribute keys store the length of the name explicitly 5019ondisk, which means that nulls are not name terminators. 5020For this section, the term "naming domain" refers to any place where names are 5021presented together -- all the names in a directory, or all the attributes of a 5022file. 5023 5024Although the Unix naming constraints are very permissive, the reality of most 5025modern-day Linux systems is that programs work with Unicode character code 5026points to support international languages. 5027These programs typically encode those code points in UTF-8 when interfacing 5028with the C library because the kernel expects null-terminated names. 5029In the common case, therefore, names found in an XFS filesystem are actually 5030UTF-8 encoded Unicode data. 5031 5032To maximize its expressiveness, the Unicode standard defines separate control 5033points for various characters that render similarly or identically in writing 5034systems around the world. 5035For example, the character "Cyrillic Small Letter A" U+0430 "а" often renders 5036identically to "Latin Small Letter A" U+0061 "a". 5037 5038The standard also permits characters to be constructed in multiple ways -- 5039either by using a defined code point, or by combining one code point with 5040various combining marks. 5041For example, the character "Angstrom Sign U+212B "Å" can also be expressed 5042as "Latin Capital Letter A" U+0041 "A" followed by "Combining Ring Above" 5043U+030A "◌̊". 5044Both sequences render identically. 5045 5046Like the standards that preceded it, Unicode also defines various control 5047characters to alter the presentation of text. 5048For example, the character "Right-to-Left Override" U+202E can trick some 5049programs into rendering "moo\\xe2\\x80\\xaegnp.txt" as "mootxt.png". 5050A second category of rendering problems involves whitespace characters. 5051If the character "Zero Width Space" U+200B is encountered in a file name, the 5052name will render identically to a name that does not have the zero width 5053space. 5054 5055If two names within a naming domain have different byte sequences but render 5056identically, a user may be confused by it. 5057The kernel, in its indifference to upper level encoding schemes, permits this. 5058Most filesystem drivers persist the byte sequence names that are given to them 5059by the VFS. 5060 5061Techniques for detecting confusable names are explained in great detail in 5062sections 4 and 5 of the 5063`Unicode Security Mechanisms <https://unicode.org/reports/tr39/>`_ 5064document. 5065When ``xfs_scrub`` detects UTF-8 encoding in use on a system, it uses the 5066Unicode normalization form NFD in conjunction with the confusable name 5067detection component of 5068`libicu <https://github.com/unicode-org/icu>`_ 5069to identify names with a directory or within a file's extended attributes that 5070could be confused for each other. 5071Names are also checked for control characters, non-rendering characters, and 5072mixing of bidirectional characters. 5073All of these potential issues are reported to the system administrator during 5074phase 5. 5075 5076Media Verification of File Data Extents 5077--------------------------------------- 5078 5079The system administrator can elect to initiate a media scan of all file data 5080blocks. 5081This scan after validation of all filesystem metadata (except for the summary 5082counters) as phase 6. 5083The scan starts by calling ``FS_IOC_GETFSMAP`` to scan the filesystem space map 5084to find areas that are allocated to file data fork extents. 5085Gaps between data fork extents that are smaller than 64k are treated as if 5086they were data fork extents to reduce the command setup overhead. 5087When the space map scan accumulates a region larger than 32MB, a media 5088verification request is sent to the disk as a directio read of the raw block 5089device. 5090 5091If the verification read fails, ``xfs_scrub`` retries with single-block reads 5092to narrow down the failure to the specific region of the media and recorded. 5093When it has finished issuing verification requests, it again uses the space 5094mapping ioctl to map the recorded media errors back to metadata structures 5095and report what has been lost. 5096For media errors in blocks owned by files, parent pointers can be used to 5097construct file paths from inode numbers for user-friendly reporting. 5098 50997. Conclusion and Future Work 5100============================= 5101 5102It is hoped that the reader of this document has followed the designs laid out 5103in this document and now has some familiarity with how XFS performs online 5104rebuilding of its metadata indices, and how filesystem users can interact with 5105that functionality. 5106Although the scope of this work is daunting, it is hoped that this guide will 5107make it easier for code readers to understand what has been built, for whom it 5108has been built, and why. 5109Please feel free to contact the XFS mailing list with questions. 5110 5111FIEXCHANGE_RANGE 5112---------------- 5113 5114As discussed earlier, a second frontend to the atomic extent swap mechanism is 5115a new ioctl call that userspace programs can use to commit updates to files 5116atomically. 5117This frontend has been out for review for several years now, though the 5118necessary refinements to online repair and lack of customer demand mean that 5119the proposal has not been pushed very hard. 5120 5121Extent Swapping with Regular User Files 5122``````````````````````````````````````` 5123 5124As mentioned earlier, XFS has long had the ability to swap extents between 5125files, which is used almost exclusively by ``xfs_fsr`` to defragment files. 5126The earliest form of this was the fork swap mechanism, where the entire 5127contents of data forks could be exchanged between two files by exchanging the 5128raw bytes in each inode fork's immediate area. 5129When XFS v5 came along with self-describing metadata, this old mechanism grew 5130some log support to continue rewriting the owner fields of BMBT blocks during 5131log recovery. 5132When the reverse mapping btree was later added to XFS, the only way to maintain 5133the consistency of the fork mappings with the reverse mapping index was to 5134develop an iterative mechanism that used deferred bmap and rmap operations to 5135swap mappings one at a time. 5136This mechanism is identical to steps 2-3 from the procedure above except for 5137the new tracking items, because the atomic extent swap mechanism is an 5138iteration of an existing mechanism and not something totally novel. 5139For the narrow case of file defragmentation, the file contents must be 5140identical, so the recovery guarantees are not much of a gain. 5141 5142Atomic extent swapping is much more flexible than the existing swapext 5143implementations because it can guarantee that the caller never sees a mix of 5144old and new contents even after a crash, and it can operate on two arbitrary 5145file fork ranges. 5146The extra flexibility enables several new use cases: 5147 5148- **Atomic commit of file writes**: A userspace process opens a file that it 5149 wants to update. 5150 Next, it opens a temporary file and calls the file clone operation to reflink 5151 the first file's contents into the temporary file. 5152 Writes to the original file should instead be written to the temporary file. 5153 Finally, the process calls the atomic extent swap system call 5154 (``FIEXCHANGE_RANGE``) to exchange the file contents, thereby committing all 5155 of the updates to the original file, or none of them. 5156 5157.. _swapext_if_unchanged: 5158 5159- **Transactional file updates**: The same mechanism as above, but the caller 5160 only wants the commit to occur if the original file's contents have not 5161 changed. 5162 To make this happen, the calling process snapshots the file modification and 5163 change timestamps of the original file before reflinking its data to the 5164 temporary file. 5165 When the program is ready to commit the changes, it passes the timestamps 5166 into the kernel as arguments to the atomic extent swap system call. 5167 The kernel only commits the changes if the provided timestamps match the 5168 original file. 5169 5170- **Emulation of atomic block device writes**: Export a block device with a 5171 logical sector size matching the filesystem block size to force all writes 5172 to be aligned to the filesystem block size. 5173 Stage all writes to a temporary file, and when that is complete, call the 5174 atomic extent swap system call with a flag to indicate that holes in the 5175 temporary file should be ignored. 5176 This emulates an atomic device write in software, and can support arbitrary 5177 scattered writes. 5178 5179Vectorized Scrub 5180---------------- 5181 5182As it turns out, the :ref:`refactoring <scrubrepair>` of repair items mentioned 5183earlier was a catalyst for enabling a vectorized scrub system call. 5184Since 2018, the cost of making a kernel call has increased considerably on some 5185systems to mitigate the effects of speculative execution attacks. 5186This incentivizes program authors to make as few system calls as possible to 5187reduce the number of times an execution path crosses a security boundary. 5188 5189With vectorized scrub, userspace pushes to the kernel the identity of a 5190filesystem object, a list of scrub types to run against that object, and a 5191simple representation of the data dependencies between the selected scrub 5192types. 5193The kernel executes as much of the caller's plan as it can until it hits a 5194dependency that cannot be satisfied due to a corruption, and tells userspace 5195how much was accomplished. 5196It is hoped that ``io_uring`` will pick up enough of this functionality that 5197online fsck can use that instead of adding a separate vectored scrub system 5198call to XFS. 5199 5200The relevant patchsets are the 5201`kernel vectorized scrub 5202<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=vectorized-scrub>`_ 5203and 5204`userspace vectorized scrub 5205<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=vectorized-scrub>`_ 5206series. 5207 5208Quality of Service Targets for Scrub 5209------------------------------------ 5210 5211One serious shortcoming of the online fsck code is that the amount of time that 5212it can spend in the kernel holding resource locks is basically unbounded. 5213Userspace is allowed to send a fatal signal to the process which will cause 5214``xfs_scrub`` to exit when it reaches a good stopping point, but there's no way 5215for userspace to provide a time budget to the kernel. 5216Given that the scrub codebase has helpers to detect fatal signals, it shouldn't 5217be too much work to allow userspace to specify a timeout for a scrub/repair 5218operation and abort the operation if it exceeds budget. 5219However, most repair functions have the property that once they begin to touch 5220ondisk metadata, the operation cannot be cancelled cleanly, after which a QoS 5221timeout is no longer useful. 5222 5223Defragmenting Free Space 5224------------------------ 5225 5226Over the years, many XFS users have requested the creation of a program to 5227clear a portion of the physical storage underlying a filesystem so that it 5228becomes a contiguous chunk of free space. 5229Call this free space defragmenter ``clearspace`` for short. 5230 5231The first piece the ``clearspace`` program needs is the ability to read the 5232reverse mapping index from userspace. 5233This already exists in the form of the ``FS_IOC_GETFSMAP`` ioctl. 5234The second piece it needs is a new fallocate mode 5235(``FALLOC_FL_MAP_FREE_SPACE``) that allocates the free space in a region and 5236maps it to a file. 5237Call this file the "space collector" file. 5238The third piece is the ability to force an online repair. 5239 5240To clear all the metadata out of a portion of physical storage, clearspace 5241uses the new fallocate map-freespace call to map any free space in that region 5242to the space collector file. 5243Next, clearspace finds all metadata blocks in that region by way of 5244``GETFSMAP`` and issues forced repair requests on the data structure. 5245This often results in the metadata being rebuilt somewhere that is not being 5246cleared. 5247After each relocation, clearspace calls the "map free space" function again to 5248collect any newly freed space in the region being cleared. 5249 5250To clear all the file data out of a portion of the physical storage, clearspace 5251uses the FSMAP information to find relevant file data blocks. 5252Having identified a good target, it uses the ``FICLONERANGE`` call on that part 5253of the file to try to share the physical space with a dummy file. 5254Cloning the extent means that the original owners cannot overwrite the 5255contents; any changes will be written somewhere else via copy-on-write. 5256Clearspace makes its own copy of the frozen extent in an area that is not being 5257cleared, and uses ``FIEDEUPRANGE`` (or the :ref:`atomic extent swap 5258<swapext_if_unchanged>` feature) to change the target file's data extent 5259mapping away from the area being cleared. 5260When all other mappings have been moved, clearspace reflinks the space into the 5261space collector file so that it becomes unavailable. 5262 5263There are further optimizations that could apply to the above algorithm. 5264To clear a piece of physical storage that has a high sharing factor, it is 5265strongly desirable to retain this sharing factor. 5266In fact, these extents should be moved first to maximize sharing factor after 5267the operation completes. 5268To make this work smoothly, clearspace needs a new ioctl 5269(``FS_IOC_GETREFCOUNTS``) to report reference count information to userspace. 5270With the refcount information exposed, clearspace can quickly find the longest, 5271most shared data extents in the filesystem, and target them first. 5272 5273**Future Work Question**: How might the filesystem move inode chunks? 5274 5275*Answer*: To move inode chunks, Dave Chinner constructed a prototype program 5276that creates a new file with the old contents and then locklessly runs around 5277the filesystem updating directory entries. 5278The operation cannot complete if the filesystem goes down. 5279That problem isn't totally insurmountable: create an inode remapping table 5280hidden behind a jump label, and a log item that tracks the kernel walking the 5281filesystem to update directory entries. 5282The trouble is, the kernel can't do anything about open files, since it cannot 5283revoke them. 5284 5285**Future Work Question**: Can static keys be used to minimize the cost of 5286supporting ``revoke()`` on XFS files? 5287 5288*Answer*: Yes. 5289Until the first revocation, the bailout code need not be in the call path at 5290all. 5291 5292The relevant patchsets are the 5293`kernel freespace defrag 5294<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/log/?h=defrag-freespace>`_ 5295and 5296`userspace freespace defrag 5297<https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=defrag-freespace>`_ 5298series. 5299 5300Shrinking Filesystems 5301--------------------- 5302 5303Removing the end of the filesystem ought to be a simple matter of evacuating 5304the data and metadata at the end of the filesystem, and handing the freed space 5305to the shrink code. 5306That requires an evacuation of the space at end of the filesystem, which is a 5307use of free space defragmentation! 5308
