xfs_trans_buf.c 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783
  1. /*
  2. * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
  3. * All Rights Reserved.
  4. *
  5. * This program is free software; you can redistribute it and/or
  6. * modify it under the terms of the GNU General Public License as
  7. * published by the Free Software Foundation.
  8. *
  9. * This program is distributed in the hope that it would be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write the Free Software Foundation,
  16. * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. #include "xfs.h"
  19. #include "xfs_fs.h"
  20. #include "xfs_types.h"
  21. #include "xfs_bit.h"
  22. #include "xfs_log.h"
  23. #include "xfs_trans.h"
  24. #include "xfs_sb.h"
  25. #include "xfs_ag.h"
  26. #include "xfs_mount.h"
  27. #include "xfs_bmap_btree.h"
  28. #include "xfs_alloc_btree.h"
  29. #include "xfs_ialloc_btree.h"
  30. #include "xfs_dinode.h"
  31. #include "xfs_inode.h"
  32. #include "xfs_buf_item.h"
  33. #include "xfs_trans_priv.h"
  34. #include "xfs_error.h"
  35. #include "xfs_trace.h"
  36. /*
  37. * Check to see if a buffer matching the given parameters is already
  38. * a part of the given transaction.
  39. */
  40. STATIC struct xfs_buf *
  41. xfs_trans_buf_item_match(
  42. struct xfs_trans *tp,
  43. struct xfs_buftarg *target,
  44. xfs_daddr_t blkno,
  45. int len)
  46. {
  47. struct xfs_log_item_desc *lidp;
  48. struct xfs_buf_log_item *blip;
  49. len = BBTOB(len);
  50. list_for_each_entry(lidp, &tp->t_items, lid_trans) {
  51. blip = (struct xfs_buf_log_item *)lidp->lid_item;
  52. if (blip->bli_item.li_type == XFS_LI_BUF &&
  53. blip->bli_buf->b_target == target &&
  54. XFS_BUF_ADDR(blip->bli_buf) == blkno &&
  55. BBTOB(blip->bli_buf->b_length) == len)
  56. return blip->bli_buf;
  57. }
  58. return NULL;
  59. }
  60. /*
  61. * Add the locked buffer to the transaction.
  62. *
  63. * The buffer must be locked, and it cannot be associated with any
  64. * transaction.
  65. *
  66. * If the buffer does not yet have a buf log item associated with it,
  67. * then allocate one for it. Then add the buf item to the transaction.
  68. */
  69. STATIC void
  70. _xfs_trans_bjoin(
  71. struct xfs_trans *tp,
  72. struct xfs_buf *bp,
  73. int reset_recur)
  74. {
  75. struct xfs_buf_log_item *bip;
  76. ASSERT(bp->b_transp == NULL);
  77. /*
  78. * The xfs_buf_log_item pointer is stored in b_fsprivate. If
  79. * it doesn't have one yet, then allocate one and initialize it.
  80. * The checks to see if one is there are in xfs_buf_item_init().
  81. */
  82. xfs_buf_item_init(bp, tp->t_mountp);
  83. bip = bp->b_fspriv;
  84. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  85. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  86. ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
  87. if (reset_recur)
  88. bip->bli_recur = 0;
  89. /*
  90. * Take a reference for this transaction on the buf item.
  91. */
  92. atomic_inc(&bip->bli_refcount);
  93. /*
  94. * Get a log_item_desc to point at the new item.
  95. */
  96. xfs_trans_add_item(tp, &bip->bli_item);
  97. /*
  98. * Initialize b_fsprivate2 so we can find it with incore_match()
  99. * in xfs_trans_get_buf() and friends above.
  100. */
  101. bp->b_transp = tp;
  102. }
  103. void
  104. xfs_trans_bjoin(
  105. struct xfs_trans *tp,
  106. struct xfs_buf *bp)
  107. {
  108. _xfs_trans_bjoin(tp, bp, 0);
  109. trace_xfs_trans_bjoin(bp->b_fspriv);
  110. }
  111. /*
  112. * Get and lock the buffer for the caller if it is not already
  113. * locked within the given transaction. If it is already locked
  114. * within the transaction, just increment its lock recursion count
  115. * and return a pointer to it.
  116. *
  117. * If the transaction pointer is NULL, make this just a normal
  118. * get_buf() call.
  119. */
  120. xfs_buf_t *
  121. xfs_trans_get_buf(xfs_trans_t *tp,
  122. xfs_buftarg_t *target_dev,
  123. xfs_daddr_t blkno,
  124. int len,
  125. uint flags)
  126. {
  127. xfs_buf_t *bp;
  128. xfs_buf_log_item_t *bip;
  129. if (flags == 0)
  130. flags = XBF_MAPPED;
  131. /*
  132. * Default to a normal get_buf() call if the tp is NULL.
  133. */
  134. if (tp == NULL)
  135. return xfs_buf_get(target_dev, blkno, len, flags);
  136. /*
  137. * If we find the buffer in the cache with this transaction
  138. * pointer in its b_fsprivate2 field, then we know we already
  139. * have it locked. In this case we just increment the lock
  140. * recursion count and return the buffer to the caller.
  141. */
  142. bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
  143. if (bp != NULL) {
  144. ASSERT(xfs_buf_islocked(bp));
  145. if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
  146. xfs_buf_stale(bp);
  147. XFS_BUF_DONE(bp);
  148. }
  149. ASSERT(bp->b_transp == tp);
  150. bip = bp->b_fspriv;
  151. ASSERT(bip != NULL);
  152. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  153. bip->bli_recur++;
  154. trace_xfs_trans_get_buf_recur(bip);
  155. return (bp);
  156. }
  157. bp = xfs_buf_get(target_dev, blkno, len, flags);
  158. if (bp == NULL) {
  159. return NULL;
  160. }
  161. ASSERT(!bp->b_error);
  162. _xfs_trans_bjoin(tp, bp, 1);
  163. trace_xfs_trans_get_buf(bp->b_fspriv);
  164. return (bp);
  165. }
  166. /*
  167. * Get and lock the superblock buffer of this file system for the
  168. * given transaction.
  169. *
  170. * We don't need to use incore_match() here, because the superblock
  171. * buffer is a private buffer which we keep a pointer to in the
  172. * mount structure.
  173. */
  174. xfs_buf_t *
  175. xfs_trans_getsb(xfs_trans_t *tp,
  176. struct xfs_mount *mp,
  177. int flags)
  178. {
  179. xfs_buf_t *bp;
  180. xfs_buf_log_item_t *bip;
  181. /*
  182. * Default to just trying to lock the superblock buffer
  183. * if tp is NULL.
  184. */
  185. if (tp == NULL) {
  186. return (xfs_getsb(mp, flags));
  187. }
  188. /*
  189. * If the superblock buffer already has this transaction
  190. * pointer in its b_fsprivate2 field, then we know we already
  191. * have it locked. In this case we just increment the lock
  192. * recursion count and return the buffer to the caller.
  193. */
  194. bp = mp->m_sb_bp;
  195. if (bp->b_transp == tp) {
  196. bip = bp->b_fspriv;
  197. ASSERT(bip != NULL);
  198. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  199. bip->bli_recur++;
  200. trace_xfs_trans_getsb_recur(bip);
  201. return (bp);
  202. }
  203. bp = xfs_getsb(mp, flags);
  204. if (bp == NULL)
  205. return NULL;
  206. _xfs_trans_bjoin(tp, bp, 1);
  207. trace_xfs_trans_getsb(bp->b_fspriv);
  208. return (bp);
  209. }
  210. #ifdef DEBUG
  211. xfs_buftarg_t *xfs_error_target;
  212. int xfs_do_error;
  213. int xfs_req_num;
  214. int xfs_error_mod = 33;
  215. #endif
  216. /*
  217. * Get and lock the buffer for the caller if it is not already
  218. * locked within the given transaction. If it has not yet been
  219. * read in, read it from disk. If it is already locked
  220. * within the transaction and already read in, just increment its
  221. * lock recursion count and return a pointer to it.
  222. *
  223. * If the transaction pointer is NULL, make this just a normal
  224. * read_buf() call.
  225. */
  226. int
  227. xfs_trans_read_buf(
  228. xfs_mount_t *mp,
  229. xfs_trans_t *tp,
  230. xfs_buftarg_t *target,
  231. xfs_daddr_t blkno,
  232. int len,
  233. uint flags,
  234. xfs_buf_t **bpp)
  235. {
  236. xfs_buf_t *bp;
  237. xfs_buf_log_item_t *bip;
  238. int error;
  239. *bpp = NULL;
  240. if (flags == 0)
  241. flags = XBF_MAPPED;
  242. /*
  243. * Default to a normal get_buf() call if the tp is NULL.
  244. */
  245. if (tp == NULL) {
  246. bp = xfs_buf_read(target, blkno, len, flags);
  247. if (!bp)
  248. return (flags & XBF_TRYLOCK) ?
  249. EAGAIN : XFS_ERROR(ENOMEM);
  250. if (bp->b_error) {
  251. error = bp->b_error;
  252. xfs_buf_ioerror_alert(bp, __func__);
  253. XFS_BUF_UNDONE(bp);
  254. xfs_buf_stale(bp);
  255. xfs_buf_relse(bp);
  256. return error;
  257. }
  258. #ifdef DEBUG
  259. if (xfs_do_error) {
  260. if (xfs_error_target == target) {
  261. if (((xfs_req_num++) % xfs_error_mod) == 0) {
  262. xfs_buf_relse(bp);
  263. xfs_debug(mp, "Returning error!");
  264. return XFS_ERROR(EIO);
  265. }
  266. }
  267. }
  268. #endif
  269. if (XFS_FORCED_SHUTDOWN(mp))
  270. goto shutdown_abort;
  271. *bpp = bp;
  272. return 0;
  273. }
  274. /*
  275. * If we find the buffer in the cache with this transaction
  276. * pointer in its b_fsprivate2 field, then we know we already
  277. * have it locked. If it is already read in we just increment
  278. * the lock recursion count and return the buffer to the caller.
  279. * If the buffer is not yet read in, then we read it in, increment
  280. * the lock recursion count, and return it to the caller.
  281. */
  282. bp = xfs_trans_buf_item_match(tp, target, blkno, len);
  283. if (bp != NULL) {
  284. ASSERT(xfs_buf_islocked(bp));
  285. ASSERT(bp->b_transp == tp);
  286. ASSERT(bp->b_fspriv != NULL);
  287. ASSERT(!bp->b_error);
  288. if (!(XFS_BUF_ISDONE(bp))) {
  289. trace_xfs_trans_read_buf_io(bp, _RET_IP_);
  290. ASSERT(!XFS_BUF_ISASYNC(bp));
  291. XFS_BUF_READ(bp);
  292. xfsbdstrat(tp->t_mountp, bp);
  293. error = xfs_buf_iowait(bp);
  294. if (error) {
  295. xfs_buf_ioerror_alert(bp, __func__);
  296. xfs_buf_relse(bp);
  297. /*
  298. * We can gracefully recover from most read
  299. * errors. Ones we can't are those that happen
  300. * after the transaction's already dirty.
  301. */
  302. if (tp->t_flags & XFS_TRANS_DIRTY)
  303. xfs_force_shutdown(tp->t_mountp,
  304. SHUTDOWN_META_IO_ERROR);
  305. return error;
  306. }
  307. }
  308. /*
  309. * We never locked this buf ourselves, so we shouldn't
  310. * brelse it either. Just get out.
  311. */
  312. if (XFS_FORCED_SHUTDOWN(mp)) {
  313. trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
  314. *bpp = NULL;
  315. return XFS_ERROR(EIO);
  316. }
  317. bip = bp->b_fspriv;
  318. bip->bli_recur++;
  319. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  320. trace_xfs_trans_read_buf_recur(bip);
  321. *bpp = bp;
  322. return 0;
  323. }
  324. bp = xfs_buf_read(target, blkno, len, flags);
  325. if (bp == NULL) {
  326. *bpp = NULL;
  327. return (flags & XBF_TRYLOCK) ?
  328. 0 : XFS_ERROR(ENOMEM);
  329. }
  330. if (bp->b_error) {
  331. error = bp->b_error;
  332. xfs_buf_stale(bp);
  333. XFS_BUF_DONE(bp);
  334. xfs_buf_ioerror_alert(bp, __func__);
  335. if (tp->t_flags & XFS_TRANS_DIRTY)
  336. xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
  337. xfs_buf_relse(bp);
  338. return error;
  339. }
  340. #ifdef DEBUG
  341. if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
  342. if (xfs_error_target == target) {
  343. if (((xfs_req_num++) % xfs_error_mod) == 0) {
  344. xfs_force_shutdown(tp->t_mountp,
  345. SHUTDOWN_META_IO_ERROR);
  346. xfs_buf_relse(bp);
  347. xfs_debug(mp, "Returning trans error!");
  348. return XFS_ERROR(EIO);
  349. }
  350. }
  351. }
  352. #endif
  353. if (XFS_FORCED_SHUTDOWN(mp))
  354. goto shutdown_abort;
  355. _xfs_trans_bjoin(tp, bp, 1);
  356. trace_xfs_trans_read_buf(bp->b_fspriv);
  357. *bpp = bp;
  358. return 0;
  359. shutdown_abort:
  360. trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
  361. xfs_buf_relse(bp);
  362. *bpp = NULL;
  363. return XFS_ERROR(EIO);
  364. }
  365. /*
  366. * Release the buffer bp which was previously acquired with one of the
  367. * xfs_trans_... buffer allocation routines if the buffer has not
  368. * been modified within this transaction. If the buffer is modified
  369. * within this transaction, do decrement the recursion count but do
  370. * not release the buffer even if the count goes to 0. If the buffer is not
  371. * modified within the transaction, decrement the recursion count and
  372. * release the buffer if the recursion count goes to 0.
  373. *
  374. * If the buffer is to be released and it was not modified before
  375. * this transaction began, then free the buf_log_item associated with it.
  376. *
  377. * If the transaction pointer is NULL, make this just a normal
  378. * brelse() call.
  379. */
  380. void
  381. xfs_trans_brelse(xfs_trans_t *tp,
  382. xfs_buf_t *bp)
  383. {
  384. xfs_buf_log_item_t *bip;
  385. /*
  386. * Default to a normal brelse() call if the tp is NULL.
  387. */
  388. if (tp == NULL) {
  389. ASSERT(bp->b_transp == NULL);
  390. xfs_buf_relse(bp);
  391. return;
  392. }
  393. ASSERT(bp->b_transp == tp);
  394. bip = bp->b_fspriv;
  395. ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
  396. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  397. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  398. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  399. trace_xfs_trans_brelse(bip);
  400. /*
  401. * If the release is just for a recursive lock,
  402. * then decrement the count and return.
  403. */
  404. if (bip->bli_recur > 0) {
  405. bip->bli_recur--;
  406. return;
  407. }
  408. /*
  409. * If the buffer is dirty within this transaction, we can't
  410. * release it until we commit.
  411. */
  412. if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
  413. return;
  414. /*
  415. * If the buffer has been invalidated, then we can't release
  416. * it until the transaction commits to disk unless it is re-dirtied
  417. * as part of this transaction. This prevents us from pulling
  418. * the item from the AIL before we should.
  419. */
  420. if (bip->bli_flags & XFS_BLI_STALE)
  421. return;
  422. ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
  423. /*
  424. * Free up the log item descriptor tracking the released item.
  425. */
  426. xfs_trans_del_item(&bip->bli_item);
  427. /*
  428. * Clear the hold flag in the buf log item if it is set.
  429. * We wouldn't want the next user of the buffer to
  430. * get confused.
  431. */
  432. if (bip->bli_flags & XFS_BLI_HOLD) {
  433. bip->bli_flags &= ~XFS_BLI_HOLD;
  434. }
  435. /*
  436. * Drop our reference to the buf log item.
  437. */
  438. atomic_dec(&bip->bli_refcount);
  439. /*
  440. * If the buf item is not tracking data in the log, then
  441. * we must free it before releasing the buffer back to the
  442. * free pool. Before releasing the buffer to the free pool,
  443. * clear the transaction pointer in b_fsprivate2 to dissolve
  444. * its relation to this transaction.
  445. */
  446. if (!xfs_buf_item_dirty(bip)) {
  447. /***
  448. ASSERT(bp->b_pincount == 0);
  449. ***/
  450. ASSERT(atomic_read(&bip->bli_refcount) == 0);
  451. ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
  452. ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
  453. xfs_buf_item_relse(bp);
  454. }
  455. bp->b_transp = NULL;
  456. xfs_buf_relse(bp);
  457. }
  458. /*
  459. * Mark the buffer as not needing to be unlocked when the buf item's
  460. * IOP_UNLOCK() routine is called. The buffer must already be locked
  461. * and associated with the given transaction.
  462. */
  463. /* ARGSUSED */
  464. void
  465. xfs_trans_bhold(xfs_trans_t *tp,
  466. xfs_buf_t *bp)
  467. {
  468. xfs_buf_log_item_t *bip = bp->b_fspriv;
  469. ASSERT(bp->b_transp == tp);
  470. ASSERT(bip != NULL);
  471. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  472. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  473. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  474. bip->bli_flags |= XFS_BLI_HOLD;
  475. trace_xfs_trans_bhold(bip);
  476. }
  477. /*
  478. * Cancel the previous buffer hold request made on this buffer
  479. * for this transaction.
  480. */
  481. void
  482. xfs_trans_bhold_release(xfs_trans_t *tp,
  483. xfs_buf_t *bp)
  484. {
  485. xfs_buf_log_item_t *bip = bp->b_fspriv;
  486. ASSERT(bp->b_transp == tp);
  487. ASSERT(bip != NULL);
  488. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  489. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  490. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  491. ASSERT(bip->bli_flags & XFS_BLI_HOLD);
  492. bip->bli_flags &= ~XFS_BLI_HOLD;
  493. trace_xfs_trans_bhold_release(bip);
  494. }
  495. /*
  496. * This is called to mark bytes first through last inclusive of the given
  497. * buffer as needing to be logged when the transaction is committed.
  498. * The buffer must already be associated with the given transaction.
  499. *
  500. * First and last are numbers relative to the beginning of this buffer,
  501. * so the first byte in the buffer is numbered 0 regardless of the
  502. * value of b_blkno.
  503. */
  504. void
  505. xfs_trans_log_buf(xfs_trans_t *tp,
  506. xfs_buf_t *bp,
  507. uint first,
  508. uint last)
  509. {
  510. xfs_buf_log_item_t *bip = bp->b_fspriv;
  511. ASSERT(bp->b_transp == tp);
  512. ASSERT(bip != NULL);
  513. ASSERT(first <= last && last < BBTOB(bp->b_length));
  514. ASSERT(bp->b_iodone == NULL ||
  515. bp->b_iodone == xfs_buf_iodone_callbacks);
  516. /*
  517. * Mark the buffer as needing to be written out eventually,
  518. * and set its iodone function to remove the buffer's buf log
  519. * item from the AIL and free it when the buffer is flushed
  520. * to disk. See xfs_buf_attach_iodone() for more details
  521. * on li_cb and xfs_buf_iodone_callbacks().
  522. * If we end up aborting this transaction, we trap this buffer
  523. * inside the b_bdstrat callback so that this won't get written to
  524. * disk.
  525. */
  526. XFS_BUF_DONE(bp);
  527. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  528. bp->b_iodone = xfs_buf_iodone_callbacks;
  529. bip->bli_item.li_cb = xfs_buf_iodone;
  530. trace_xfs_trans_log_buf(bip);
  531. /*
  532. * If we invalidated the buffer within this transaction, then
  533. * cancel the invalidation now that we're dirtying the buffer
  534. * again. There are no races with the code in xfs_buf_item_unpin(),
  535. * because we have a reference to the buffer this entire time.
  536. */
  537. if (bip->bli_flags & XFS_BLI_STALE) {
  538. bip->bli_flags &= ~XFS_BLI_STALE;
  539. ASSERT(XFS_BUF_ISSTALE(bp));
  540. XFS_BUF_UNSTALE(bp);
  541. bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
  542. }
  543. tp->t_flags |= XFS_TRANS_DIRTY;
  544. bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
  545. bip->bli_flags |= XFS_BLI_LOGGED;
  546. xfs_buf_item_log(bip, first, last);
  547. }
  548. /*
  549. * Invalidate a buffer that is being used within a transaction.
  550. *
  551. * Typically this is because the blocks in the buffer are being freed, so we
  552. * need to prevent it from being written out when we're done. Allowing it
  553. * to be written again might overwrite data in the free blocks if they are
  554. * reallocated to a file.
  555. *
  556. * We prevent the buffer from being written out by marking it stale. We can't
  557. * get rid of the buf log item at this point because the buffer may still be
  558. * pinned by another transaction. If that is the case, then we'll wait until
  559. * the buffer is committed to disk for the last time (we can tell by the ref
  560. * count) and free it in xfs_buf_item_unpin(). Until that happens we will
  561. * keep the buffer locked so that the buffer and buf log item are not reused.
  562. *
  563. * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
  564. * the buf item. This will be used at recovery time to determine that copies
  565. * of the buffer in the log before this should not be replayed.
  566. *
  567. * We mark the item descriptor and the transaction dirty so that we'll hold
  568. * the buffer until after the commit.
  569. *
  570. * Since we're invalidating the buffer, we also clear the state about which
  571. * parts of the buffer have been logged. We also clear the flag indicating
  572. * that this is an inode buffer since the data in the buffer will no longer
  573. * be valid.
  574. *
  575. * We set the stale bit in the buffer as well since we're getting rid of it.
  576. */
  577. void
  578. xfs_trans_binval(
  579. xfs_trans_t *tp,
  580. xfs_buf_t *bp)
  581. {
  582. xfs_buf_log_item_t *bip = bp->b_fspriv;
  583. ASSERT(bp->b_transp == tp);
  584. ASSERT(bip != NULL);
  585. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  586. trace_xfs_trans_binval(bip);
  587. if (bip->bli_flags & XFS_BLI_STALE) {
  588. /*
  589. * If the buffer is already invalidated, then
  590. * just return.
  591. */
  592. ASSERT(XFS_BUF_ISSTALE(bp));
  593. ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
  594. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
  595. ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
  596. ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
  597. ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
  598. return;
  599. }
  600. xfs_buf_stale(bp);
  601. bip->bli_flags |= XFS_BLI_STALE;
  602. bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
  603. bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
  604. bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
  605. memset((char *)(bip->bli_format.blf_data_map), 0,
  606. (bip->bli_format.blf_map_size * sizeof(uint)));
  607. bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
  608. tp->t_flags |= XFS_TRANS_DIRTY;
  609. }
  610. /*
  611. * This call is used to indicate that the buffer contains on-disk inodes which
  612. * must be handled specially during recovery. They require special handling
  613. * because only the di_next_unlinked from the inodes in the buffer should be
  614. * recovered. The rest of the data in the buffer is logged via the inodes
  615. * themselves.
  616. *
  617. * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
  618. * transferred to the buffer's log format structure so that we'll know what to
  619. * do at recovery time.
  620. */
  621. void
  622. xfs_trans_inode_buf(
  623. xfs_trans_t *tp,
  624. xfs_buf_t *bp)
  625. {
  626. xfs_buf_log_item_t *bip = bp->b_fspriv;
  627. ASSERT(bp->b_transp == tp);
  628. ASSERT(bip != NULL);
  629. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  630. bip->bli_flags |= XFS_BLI_INODE_BUF;
  631. }
  632. /*
  633. * This call is used to indicate that the buffer is going to
  634. * be staled and was an inode buffer. This means it gets
  635. * special processing during unpin - where any inodes
  636. * associated with the buffer should be removed from ail.
  637. * There is also special processing during recovery,
  638. * any replay of the inodes in the buffer needs to be
  639. * prevented as the buffer may have been reused.
  640. */
  641. void
  642. xfs_trans_stale_inode_buf(
  643. xfs_trans_t *tp,
  644. xfs_buf_t *bp)
  645. {
  646. xfs_buf_log_item_t *bip = bp->b_fspriv;
  647. ASSERT(bp->b_transp == tp);
  648. ASSERT(bip != NULL);
  649. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  650. bip->bli_flags |= XFS_BLI_STALE_INODE;
  651. bip->bli_item.li_cb = xfs_buf_iodone;
  652. }
  653. /*
  654. * Mark the buffer as being one which contains newly allocated
  655. * inodes. We need to make sure that even if this buffer is
  656. * relogged as an 'inode buf' we still recover all of the inode
  657. * images in the face of a crash. This works in coordination with
  658. * xfs_buf_item_committed() to ensure that the buffer remains in the
  659. * AIL at its original location even after it has been relogged.
  660. */
  661. /* ARGSUSED */
  662. void
  663. xfs_trans_inode_alloc_buf(
  664. xfs_trans_t *tp,
  665. xfs_buf_t *bp)
  666. {
  667. xfs_buf_log_item_t *bip = bp->b_fspriv;
  668. ASSERT(bp->b_transp == tp);
  669. ASSERT(bip != NULL);
  670. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  671. bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
  672. }
  673. /*
  674. * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
  675. * dquots. However, unlike in inode buffer recovery, dquot buffers get
  676. * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
  677. * The only thing that makes dquot buffers different from regular
  678. * buffers is that we must not replay dquot bufs when recovering
  679. * if a _corresponding_ quotaoff has happened. We also have to distinguish
  680. * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
  681. * can be turned off independently.
  682. */
  683. /* ARGSUSED */
  684. void
  685. xfs_trans_dquot_buf(
  686. xfs_trans_t *tp,
  687. xfs_buf_t *bp,
  688. uint type)
  689. {
  690. xfs_buf_log_item_t *bip = bp->b_fspriv;
  691. ASSERT(bp->b_transp == tp);
  692. ASSERT(bip != NULL);
  693. ASSERT(type == XFS_BLF_UDQUOT_BUF ||
  694. type == XFS_BLF_PDQUOT_BUF ||
  695. type == XFS_BLF_GDQUOT_BUF);
  696. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  697. bip->bli_format.blf_flags |= type;
  698. }